xref: /dflybsd-src/sys/kern/vfs_syscalls.c (revision 3cf8dfbcc5e851c3571a9caa420ccd50eb89a824)
1 /*
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)vfs_syscalls.c	8.13 (Berkeley) 4/15/94
39  * $FreeBSD: src/sys/kern/vfs_syscalls.c,v 1.151.2.18 2003/04/04 20:35:58 tegge Exp $
40  * $DragonFly: src/sys/kern/vfs_syscalls.c,v 1.135 2008/11/11 00:55:49 pavalos Exp $
41  */
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/buf.h>
46 #include <sys/conf.h>
47 #include <sys/sysent.h>
48 #include <sys/malloc.h>
49 #include <sys/mount.h>
50 #include <sys/mountctl.h>
51 #include <sys/sysproto.h>
52 #include <sys/filedesc.h>
53 #include <sys/kernel.h>
54 #include <sys/fcntl.h>
55 #include <sys/file.h>
56 #include <sys/linker.h>
57 #include <sys/stat.h>
58 #include <sys/unistd.h>
59 #include <sys/vnode.h>
60 #include <sys/proc.h>
61 #include <sys/namei.h>
62 #include <sys/nlookup.h>
63 #include <sys/dirent.h>
64 #include <sys/extattr.h>
65 #include <sys/spinlock.h>
66 #include <sys/kern_syscall.h>
67 #include <sys/objcache.h>
68 #include <sys/sysctl.h>
69 
70 #include <sys/buf2.h>
71 #include <sys/file2.h>
72 #include <sys/spinlock2.h>
73 
74 #include <vm/vm.h>
75 #include <vm/vm_object.h>
76 #include <vm/vm_page.h>
77 
78 #include <machine/limits.h>
79 #include <machine/stdarg.h>
80 
81 #include <vfs/union/union.h>
82 
83 static void mount_warning(struct mount *mp, const char *ctl, ...);
84 static int mount_path(struct proc *p, struct mount *mp, char **rb, char **fb);
85 static int checkvp_chdir (struct vnode *vn, struct thread *td);
86 static void checkdirs (struct nchandle *old_nch, struct nchandle *new_nch);
87 static int chroot_refuse_vdir_fds (struct filedesc *fdp);
88 static int chroot_visible_mnt(struct mount *mp, struct proc *p);
89 static int getutimes (const struct timeval *, struct timespec *);
90 static int setfown (struct vnode *, uid_t, gid_t);
91 static int setfmode (struct vnode *, int);
92 static int setfflags (struct vnode *, int);
93 static int setutimes (struct vnode *, const struct timespec *, int);
94 static int	usermount = 0;	/* if 1, non-root can mount fs. */
95 
96 int (*union_dircheckp) (struct thread *, struct vnode **, struct file *);
97 
98 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, "");
99 
100 /*
101  * Virtual File System System Calls
102  */
103 
104 /*
105  * Mount a file system.
106  */
107 /*
108  * mount_args(char *type, char *path, int flags, caddr_t data)
109  */
110 /* ARGSUSED */
111 int
112 sys_mount(struct mount_args *uap)
113 {
114 	struct thread *td = curthread;
115 	struct proc *p = td->td_proc;
116 	struct vnode *vp;
117 	struct nchandle nch;
118 	struct mount *mp;
119 	struct vfsconf *vfsp;
120 	int error, flag = 0, flag2 = 0;
121 	int hasmount;
122 	struct vattr va;
123 	struct nlookupdata nd;
124 	char fstypename[MFSNAMELEN];
125 	struct ucred *cred = p->p_ucred;
126 
127 	KKASSERT(p);
128 	if (cred->cr_prison != NULL)
129 		return (EPERM);
130 	if (usermount == 0 && (error = suser(td)))
131 		return (error);
132 	/*
133 	 * Do not allow NFS export by non-root users.
134 	 */
135 	if (uap->flags & MNT_EXPORTED) {
136 		error = suser(td);
137 		if (error)
138 			return (error);
139 	}
140 	/*
141 	 * Silently enforce MNT_NOSUID and MNT_NODEV for non-root users
142 	 */
143 	if (suser(td))
144 		uap->flags |= MNT_NOSUID | MNT_NODEV;
145 
146 	/*
147 	 * Lookup the requested path and extract the nch and vnode.
148 	 */
149 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
150 	if (error == 0) {
151 		if ((error = nlookup(&nd)) == 0) {
152 			if (nd.nl_nch.ncp->nc_vp == NULL)
153 				error = ENOENT;
154 		}
155 	}
156 	if (error) {
157 		nlookup_done(&nd);
158 		return (error);
159 	}
160 
161 	/*
162 	 * Extract the locked+refd ncp and cleanup the nd structure
163 	 */
164 	nch = nd.nl_nch;
165 	cache_zero(&nd.nl_nch);
166 	nlookup_done(&nd);
167 
168 	if ((nch.ncp->nc_flag & NCF_ISMOUNTPT) && cache_findmount(&nch))
169 		hasmount = 1;
170 	else
171 		hasmount = 0;
172 
173 
174 	/*
175 	 * now we have the locked ref'd nch and unreferenced vnode.
176 	 */
177 	vp = nch.ncp->nc_vp;
178 	if ((error = vget(vp, LK_EXCLUSIVE)) != 0) {
179 		cache_put(&nch);
180 		return (error);
181 	}
182 	cache_unlock(&nch);
183 
184 	/*
185 	 * Now we have an unlocked ref'd nch and a locked ref'd vp
186 	 */
187 	if (uap->flags & MNT_UPDATE) {
188 		if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) {
189 			cache_drop(&nch);
190 			vput(vp);
191 			return (EINVAL);
192 		}
193 		mp = vp->v_mount;
194 		flag = mp->mnt_flag;
195 		flag2 = mp->mnt_kern_flag;
196 		/*
197 		 * We only allow the filesystem to be reloaded if it
198 		 * is currently mounted read-only.
199 		 */
200 		if ((uap->flags & MNT_RELOAD) &&
201 		    ((mp->mnt_flag & MNT_RDONLY) == 0)) {
202 			cache_drop(&nch);
203 			vput(vp);
204 			return (EOPNOTSUPP);	/* Needs translation */
205 		}
206 		/*
207 		 * Only root, or the user that did the original mount is
208 		 * permitted to update it.
209 		 */
210 		if (mp->mnt_stat.f_owner != cred->cr_uid &&
211 		    (error = suser(td))) {
212 			cache_drop(&nch);
213 			vput(vp);
214 			return (error);
215 		}
216 		if (vfs_busy(mp, LK_NOWAIT)) {
217 			cache_drop(&nch);
218 			vput(vp);
219 			return (EBUSY);
220 		}
221 		if ((vp->v_flag & VMOUNT) != 0 || hasmount) {
222 			cache_drop(&nch);
223 			vfs_unbusy(mp);
224 			vput(vp);
225 			return (EBUSY);
226 		}
227 		vp->v_flag |= VMOUNT;
228 		mp->mnt_flag |=
229 		    uap->flags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE);
230 		vn_unlock(vp);
231 		goto update;
232 	}
233 	/*
234 	 * If the user is not root, ensure that they own the directory
235 	 * onto which we are attempting to mount.
236 	 */
237 	if ((error = VOP_GETATTR(vp, &va)) ||
238 	    (va.va_uid != cred->cr_uid && (error = suser(td)))) {
239 		cache_drop(&nch);
240 		vput(vp);
241 		return (error);
242 	}
243 	if ((error = vinvalbuf(vp, V_SAVE, 0, 0)) != 0) {
244 		cache_drop(&nch);
245 		vput(vp);
246 		return (error);
247 	}
248 	if (vp->v_type != VDIR) {
249 		cache_drop(&nch);
250 		vput(vp);
251 		return (ENOTDIR);
252 	}
253 	if (vp->v_mount->mnt_kern_flag & MNTK_NOSTKMNT) {
254 		cache_drop(&nch);
255 		vput(vp);
256 		return (EPERM);
257 	}
258 	if ((error = copyinstr(uap->type, fstypename, MFSNAMELEN, NULL)) != 0) {
259 		cache_drop(&nch);
260 		vput(vp);
261 		return (error);
262 	}
263 	vfsp = vfsconf_find_by_name(fstypename);
264 	if (vfsp == NULL) {
265 		linker_file_t lf;
266 
267 		/* Only load modules for root (very important!) */
268 		if ((error = suser(td)) != 0) {
269 			cache_drop(&nch);
270 			vput(vp);
271 			return error;
272 		}
273 		error = linker_load_file(fstypename, &lf);
274 		if (error || lf == NULL) {
275 			cache_drop(&nch);
276 			vput(vp);
277 			if (lf == NULL)
278 				error = ENODEV;
279 			return error;
280 		}
281 		lf->userrefs++;
282 		/* lookup again, see if the VFS was loaded */
283 		vfsp = vfsconf_find_by_name(fstypename);
284 		if (vfsp == NULL) {
285 			lf->userrefs--;
286 			linker_file_unload(lf);
287 			cache_drop(&nch);
288 			vput(vp);
289 			return (ENODEV);
290 		}
291 	}
292 	if ((vp->v_flag & VMOUNT) != 0 || hasmount) {
293 		cache_drop(&nch);
294 		vput(vp);
295 		return (EBUSY);
296 	}
297 	vp->v_flag |= VMOUNT;
298 
299 	/*
300 	 * Allocate and initialize the filesystem.
301 	 */
302 	mp = kmalloc(sizeof(struct mount), M_MOUNT, M_ZERO|M_WAITOK);
303 	TAILQ_INIT(&mp->mnt_nvnodelist);
304 	TAILQ_INIT(&mp->mnt_reservedvnlist);
305 	TAILQ_INIT(&mp->mnt_jlist);
306 	mp->mnt_nvnodelistsize = 0;
307 	lockinit(&mp->mnt_lock, "vfslock", 0, 0);
308 	vfs_busy(mp, LK_NOWAIT);
309 	mp->mnt_op = vfsp->vfc_vfsops;
310 	mp->mnt_vfc = vfsp;
311 	vfsp->vfc_refcount++;
312 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
313 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
314 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
315 	mp->mnt_stat.f_owner = cred->cr_uid;
316 	mp->mnt_iosize_max = DFLTPHYS;
317 	vn_unlock(vp);
318 update:
319 	/*
320 	 * Set the mount level flags.
321 	 */
322 	if (uap->flags & MNT_RDONLY)
323 		mp->mnt_flag |= MNT_RDONLY;
324 	else if (mp->mnt_flag & MNT_RDONLY)
325 		mp->mnt_kern_flag |= MNTK_WANTRDWR;
326 	mp->mnt_flag &=~ (MNT_NOSUID | MNT_NOEXEC | MNT_NODEV |
327 	    MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_NOATIME |
328 	    MNT_NOSYMFOLLOW | MNT_IGNORE |
329 	    MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR);
330 	mp->mnt_flag |= uap->flags & (MNT_NOSUID | MNT_NOEXEC |
331 	    MNT_NODEV | MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_FORCE |
332 	    MNT_NOSYMFOLLOW | MNT_IGNORE |
333 	    MNT_NOATIME | MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR);
334 	/*
335 	 * Mount the filesystem.
336 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
337 	 * get.
338 	 */
339 	error = VFS_MOUNT(mp, uap->path, uap->data, cred);
340 	if (mp->mnt_flag & MNT_UPDATE) {
341 		if (mp->mnt_kern_flag & MNTK_WANTRDWR)
342 			mp->mnt_flag &= ~MNT_RDONLY;
343 		mp->mnt_flag &=~ (MNT_UPDATE | MNT_RELOAD | MNT_FORCE);
344 		mp->mnt_kern_flag &=~ MNTK_WANTRDWR;
345 		if (error) {
346 			mp->mnt_flag = flag;
347 			mp->mnt_kern_flag = flag2;
348 		}
349 		vfs_unbusy(mp);
350 		vp->v_flag &= ~VMOUNT;
351 		vrele(vp);
352 		cache_drop(&nch);
353 		return (error);
354 	}
355 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
356 	/*
357 	 * Put the new filesystem on the mount list after root.  The mount
358 	 * point gets its own mnt_ncmountpt (unless the VFS already set one
359 	 * up) which represents the root of the mount.  The lookup code
360 	 * detects the mount point going forward and checks the root of
361 	 * the mount going backwards.
362 	 *
363 	 * It is not necessary to invalidate or purge the vnode underneath
364 	 * because elements under the mount will be given their own glue
365 	 * namecache record.
366 	 */
367 	if (!error) {
368 		if (mp->mnt_ncmountpt.ncp == NULL) {
369 			/*
370 			 * allocate, then unlock, but leave the ref intact
371 			 */
372 			cache_allocroot(&mp->mnt_ncmountpt, mp, NULL);
373 			cache_unlock(&mp->mnt_ncmountpt);
374 		}
375 		mp->mnt_ncmounton = nch;		/* inherits ref */
376 		nch.ncp->nc_flag |= NCF_ISMOUNTPT;
377 
378 		/* XXX get the root of the fs and cache_setvp(mnt_ncmountpt...) */
379 		vp->v_flag &= ~VMOUNT;
380 		mountlist_insert(mp, MNTINS_LAST);
381 		checkdirs(&mp->mnt_ncmounton, &mp->mnt_ncmountpt);
382 		vn_unlock(vp);
383 		error = vfs_allocate_syncvnode(mp);
384 		vfs_unbusy(mp);
385 		error = VFS_START(mp, 0);
386 		vrele(vp);
387 	} else {
388 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
389 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
390 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
391 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
392 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
393 		vp->v_flag &= ~VMOUNT;
394 		mp->mnt_vfc->vfc_refcount--;
395 		vfs_unbusy(mp);
396 		kfree(mp, M_MOUNT);
397 		cache_drop(&nch);
398 		vput(vp);
399 	}
400 	return (error);
401 }
402 
403 /*
404  * Scan all active processes to see if any of them have a current
405  * or root directory onto which the new filesystem has just been
406  * mounted. If so, replace them with the new mount point.
407  *
408  * The passed ncp is ref'd and locked (from the mount code) and
409  * must be associated with the vnode representing the root of the
410  * mount point.
411  */
412 struct checkdirs_info {
413 	struct nchandle old_nch;
414 	struct nchandle new_nch;
415 	struct vnode *old_vp;
416 	struct vnode *new_vp;
417 };
418 
419 static int checkdirs_callback(struct proc *p, void *data);
420 
421 static void
422 checkdirs(struct nchandle *old_nch, struct nchandle *new_nch)
423 {
424 	struct checkdirs_info info;
425 	struct vnode *olddp;
426 	struct vnode *newdp;
427 	struct mount *mp;
428 
429 	/*
430 	 * If the old mount point's vnode has a usecount of 1, it is not
431 	 * being held as a descriptor anywhere.
432 	 */
433 	olddp = old_nch->ncp->nc_vp;
434 	if (olddp == NULL || olddp->v_sysref.refcnt == 1)
435 		return;
436 
437 	/*
438 	 * Force the root vnode of the new mount point to be resolved
439 	 * so we can update any matching processes.
440 	 */
441 	mp = new_nch->mount;
442 	if (VFS_ROOT(mp, &newdp))
443 		panic("mount: lost mount");
444 	cache_setunresolved(new_nch);
445 	cache_setvp(new_nch, newdp);
446 
447 	/*
448 	 * Special handling of the root node
449 	 */
450 	if (rootvnode == olddp) {
451 		vref(newdp);
452 		vfs_cache_setroot(newdp, cache_hold(new_nch));
453 	}
454 
455 	/*
456 	 * Pass newdp separately so the callback does not have to access
457 	 * it via new_nch->ncp->nc_vp.
458 	 */
459 	info.old_nch = *old_nch;
460 	info.new_nch = *new_nch;
461 	info.new_vp = newdp;
462 	allproc_scan(checkdirs_callback, &info);
463 	vput(newdp);
464 }
465 
466 /*
467  * NOTE: callback is not MP safe because the scanned process's filedesc
468  * structure can be ripped out from under us, amoung other things.
469  */
470 static int
471 checkdirs_callback(struct proc *p, void *data)
472 {
473 	struct checkdirs_info *info = data;
474 	struct filedesc *fdp;
475 	struct nchandle ncdrop1;
476 	struct nchandle ncdrop2;
477 	struct vnode *vprele1;
478 	struct vnode *vprele2;
479 
480 	if ((fdp = p->p_fd) != NULL) {
481 		cache_zero(&ncdrop1);
482 		cache_zero(&ncdrop2);
483 		vprele1 = NULL;
484 		vprele2 = NULL;
485 
486 		/*
487 		 * MPUNSAFE - XXX fdp can be pulled out from under a
488 		 * foreign process.
489 		 *
490 		 * A shared filedesc is ok, we don't have to copy it
491 		 * because we are making this change globally.
492 		 */
493 		spin_lock_wr(&fdp->fd_spin);
494 		if (fdp->fd_ncdir.mount == info->old_nch.mount &&
495 		    fdp->fd_ncdir.ncp == info->old_nch.ncp) {
496 			vprele1 = fdp->fd_cdir;
497 			vref(info->new_vp);
498 			fdp->fd_cdir = info->new_vp;
499 			ncdrop1 = fdp->fd_ncdir;
500 			cache_copy(&info->new_nch, &fdp->fd_ncdir);
501 		}
502 		if (fdp->fd_nrdir.mount == info->old_nch.mount &&
503 		    fdp->fd_nrdir.ncp == info->old_nch.ncp) {
504 			vprele2 = fdp->fd_rdir;
505 			vref(info->new_vp);
506 			fdp->fd_rdir = info->new_vp;
507 			ncdrop2 = fdp->fd_nrdir;
508 			cache_copy(&info->new_nch, &fdp->fd_nrdir);
509 		}
510 		spin_unlock_wr(&fdp->fd_spin);
511 		if (ncdrop1.ncp)
512 			cache_drop(&ncdrop1);
513 		if (ncdrop2.ncp)
514 			cache_drop(&ncdrop2);
515 		if (vprele1)
516 			vrele(vprele1);
517 		if (vprele2)
518 			vrele(vprele2);
519 	}
520 	return(0);
521 }
522 
523 /*
524  * Unmount a file system.
525  *
526  * Note: unmount takes a path to the vnode mounted on as argument,
527  * not special file (as before).
528  */
529 /*
530  * umount_args(char *path, int flags)
531  */
532 /* ARGSUSED */
533 int
534 sys_unmount(struct unmount_args *uap)
535 {
536 	struct thread *td = curthread;
537 	struct proc *p = td->td_proc;
538 	struct mount *mp = NULL;
539 	int error;
540 	struct nlookupdata nd;
541 
542 	KKASSERT(p);
543 	if (p->p_ucred->cr_prison != NULL)
544 		return (EPERM);
545 	if (usermount == 0 && (error = suser(td)))
546 		return (error);
547 
548 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
549 	if (error == 0)
550 		error = nlookup(&nd);
551 	if (error)
552 		goto out;
553 
554 	mp = nd.nl_nch.mount;
555 
556 	/*
557 	 * Only root, or the user that did the original mount is
558 	 * permitted to unmount this filesystem.
559 	 */
560 	if ((mp->mnt_stat.f_owner != p->p_ucred->cr_uid) &&
561 	    (error = suser(td)))
562 		goto out;
563 
564 	/*
565 	 * Don't allow unmounting the root file system.
566 	 */
567 	if (mp->mnt_flag & MNT_ROOTFS) {
568 		error = EINVAL;
569 		goto out;
570 	}
571 
572 	/*
573 	 * Must be the root of the filesystem
574 	 */
575 	if (nd.nl_nch.ncp != mp->mnt_ncmountpt.ncp) {
576 		error = EINVAL;
577 		goto out;
578 	}
579 
580 out:
581 	nlookup_done(&nd);
582 	if (error)
583 		return (error);
584 	return (dounmount(mp, uap->flags));
585 }
586 
587 /*
588  * Do the actual file system unmount.
589  */
590 static int
591 dounmount_interlock(struct mount *mp)
592 {
593 	if (mp->mnt_kern_flag & MNTK_UNMOUNT)
594 		return (EBUSY);
595 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
596 	return(0);
597 }
598 
599 int
600 dounmount(struct mount *mp, int flags)
601 {
602 	struct namecache *ncp;
603 	struct nchandle nch;
604 	struct vnode *vp;
605 	int error;
606 	int async_flag;
607 	int lflags;
608 	int freeok = 1;
609 
610 	/*
611 	 * Exclusive access for unmounting purposes
612 	 */
613 	if ((error = mountlist_interlock(dounmount_interlock, mp)) != 0)
614 		return (error);
615 
616 	/*
617 	 * Allow filesystems to detect that a forced unmount is in progress.
618 	 */
619 	if (flags & MNT_FORCE)
620 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
621 	lflags = LK_EXCLUSIVE | ((flags & MNT_FORCE) ? 0 : LK_NOWAIT);
622 	error = lockmgr(&mp->mnt_lock, lflags);
623 	if (error) {
624 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
625 		if (mp->mnt_kern_flag & MNTK_MWAIT)
626 			wakeup(mp);
627 		return (error);
628 	}
629 
630 	if (mp->mnt_flag & MNT_EXPUBLIC)
631 		vfs_setpublicfs(NULL, NULL, NULL);
632 
633 	vfs_msync(mp, MNT_WAIT);
634 	async_flag = mp->mnt_flag & MNT_ASYNC;
635 	mp->mnt_flag &=~ MNT_ASYNC;
636 
637 	/*
638 	 * If this filesystem isn't aliasing other filesystems,
639 	 * try to invalidate any remaining namecache entries and
640 	 * check the count afterwords.
641 	 */
642 	if ((mp->mnt_kern_flag & MNTK_NCALIASED) == 0) {
643 		cache_lock(&mp->mnt_ncmountpt);
644 		cache_inval(&mp->mnt_ncmountpt, CINV_DESTROY|CINV_CHILDREN);
645 		cache_unlock(&mp->mnt_ncmountpt);
646 
647 		if ((ncp = mp->mnt_ncmountpt.ncp) != NULL &&
648 		    (ncp->nc_refs != 1 || TAILQ_FIRST(&ncp->nc_list))) {
649 
650 			if ((flags & MNT_FORCE) == 0) {
651 				error = EBUSY;
652 				mount_warning(mp, "Cannot unmount: "
653 						  "%d namecache "
654 						  "references still "
655 						  "present",
656 						  ncp->nc_refs - 1);
657 			} else {
658 				mount_warning(mp, "Forced unmount: "
659 						  "%d namecache "
660 						  "references still "
661 						  "present",
662 						  ncp->nc_refs - 1);
663 				freeok = 0;
664 			}
665 		}
666 	}
667 
668 	/*
669 	 * nchandle records ref the mount structure.  Expect a count of 1
670 	 * (our mount->mnt_ncmountpt).
671 	 */
672 	if (mp->mnt_refs != 1) {
673 		if ((flags & MNT_FORCE) == 0) {
674 			mount_warning(mp, "Cannot unmount: "
675 					  "%d process references still "
676 					  "present", mp->mnt_refs);
677 			error = EBUSY;
678 		} else {
679 			mount_warning(mp, "Forced unmount: "
680 					  "%d process references still "
681 					  "present", mp->mnt_refs);
682 			freeok = 0;
683 		}
684 	}
685 
686 	/*
687 	 * Decomission our special mnt_syncer vnode.  This also stops
688 	 * the vnlru code.  If we are unable to unmount we recommission
689 	 * the vnode.
690 	 */
691 	if (error == 0) {
692 		if ((vp = mp->mnt_syncer) != NULL) {
693 			mp->mnt_syncer = NULL;
694 			vrele(vp);
695 		}
696 		if (((mp->mnt_flag & MNT_RDONLY) ||
697 		     (error = VFS_SYNC(mp, MNT_WAIT)) == 0) ||
698 		    (flags & MNT_FORCE)) {
699 			error = VFS_UNMOUNT(mp, flags);
700 		}
701 	}
702 	if (error) {
703 		if (mp->mnt_syncer == NULL)
704 			vfs_allocate_syncvnode(mp);
705 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
706 		mp->mnt_flag |= async_flag;
707 		lockmgr(&mp->mnt_lock, LK_RELEASE);
708 		if (mp->mnt_kern_flag & MNTK_MWAIT)
709 			wakeup(mp);
710 		return (error);
711 	}
712 	/*
713 	 * Clean up any journals still associated with the mount after
714 	 * filesystem activity has ceased.
715 	 */
716 	journal_remove_all_journals(mp,
717 	    ((flags & MNT_FORCE) ? MC_JOURNAL_STOP_IMM : 0));
718 
719 	mountlist_remove(mp);
720 
721 	/*
722 	 * Remove any installed vnode ops here so the individual VFSs don't
723 	 * have to.
724 	 */
725 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
726 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
727 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
728 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
729 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
730 
731 	if (mp->mnt_ncmountpt.ncp != NULL) {
732 		nch = mp->mnt_ncmountpt;
733 		cache_zero(&mp->mnt_ncmountpt);
734 		cache_clrmountpt(&nch);
735 		cache_drop(&nch);
736 	}
737 	if (mp->mnt_ncmounton.ncp != NULL) {
738 		nch = mp->mnt_ncmounton;
739 		cache_zero(&mp->mnt_ncmounton);
740 		cache_clrmountpt(&nch);
741 		cache_drop(&nch);
742 	}
743 
744 	mp->mnt_vfc->vfc_refcount--;
745 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist))
746 		panic("unmount: dangling vnode");
747 	lockmgr(&mp->mnt_lock, LK_RELEASE);
748 	if (mp->mnt_kern_flag & MNTK_MWAIT)
749 		wakeup(mp);
750 	if (freeok)
751 		kfree(mp, M_MOUNT);
752 	return (0);
753 }
754 
755 static
756 void
757 mount_warning(struct mount *mp, const char *ctl, ...)
758 {
759 	char *ptr;
760 	char *buf;
761 	__va_list va;
762 
763 	__va_start(va, ctl);
764 	if (cache_fullpath(NULL, &mp->mnt_ncmounton, &ptr, &buf) == 0) {
765 		kprintf("unmount(%s): ", ptr);
766 		kvprintf(ctl, va);
767 		kprintf("\n");
768 		kfree(buf, M_TEMP);
769 	} else {
770 		kprintf("unmount(%p", mp);
771 		if (mp->mnt_ncmounton.ncp && mp->mnt_ncmounton.ncp->nc_name)
772 			kprintf(",%s", mp->mnt_ncmounton.ncp->nc_name);
773 		kprintf("): ");
774 		kvprintf(ctl, va);
775 		kprintf("\n");
776 	}
777 	__va_end(va);
778 }
779 
780 /*
781  * Shim cache_fullpath() to handle the case where a process is chrooted into
782  * a subdirectory of a mount.  In this case if the root mount matches the
783  * process root directory's mount we have to specify the process's root
784  * directory instead of the mount point, because the mount point might
785  * be above the root directory.
786  */
787 static
788 int
789 mount_path(struct proc *p, struct mount *mp, char **rb, char **fb)
790 {
791 	struct nchandle *nch;
792 
793 	if (p && p->p_fd->fd_nrdir.mount == mp)
794 		nch = &p->p_fd->fd_nrdir;
795 	else
796 		nch = &mp->mnt_ncmountpt;
797 	return(cache_fullpath(p, nch, rb, fb));
798 }
799 
800 /*
801  * Sync each mounted filesystem.
802  */
803 
804 #ifdef DEBUG
805 static int syncprt = 0;
806 SYSCTL_INT(_debug, OID_AUTO, syncprt, CTLFLAG_RW, &syncprt, 0, "");
807 #endif /* DEBUG */
808 
809 static int sync_callback(struct mount *mp, void *data);
810 
811 /* ARGSUSED */
812 int
813 sys_sync(struct sync_args *uap)
814 {
815 	mountlist_scan(sync_callback, NULL, MNTSCAN_FORWARD);
816 #ifdef DEBUG
817 	/*
818 	 * print out buffer pool stat information on each sync() call.
819 	 */
820 	if (syncprt)
821 		vfs_bufstats();
822 #endif /* DEBUG */
823 	return (0);
824 }
825 
826 static
827 int
828 sync_callback(struct mount *mp, void *data __unused)
829 {
830 	int asyncflag;
831 
832 	if ((mp->mnt_flag & MNT_RDONLY) == 0) {
833 		asyncflag = mp->mnt_flag & MNT_ASYNC;
834 		mp->mnt_flag &= ~MNT_ASYNC;
835 		vfs_msync(mp, MNT_NOWAIT);
836 		VFS_SYNC(mp, MNT_NOWAIT);
837 		mp->mnt_flag |= asyncflag;
838 	}
839 	return(0);
840 }
841 
842 /* XXX PRISON: could be per prison flag */
843 static int prison_quotas;
844 #if 0
845 SYSCTL_INT(_kern_prison, OID_AUTO, quotas, CTLFLAG_RW, &prison_quotas, 0, "");
846 #endif
847 
848 /*
849  *  quotactl_args(char *path, int fcmd, int uid, caddr_t arg)
850  *
851  * Change filesystem quotas.
852  */
853 /* ARGSUSED */
854 int
855 sys_quotactl(struct quotactl_args *uap)
856 {
857 	struct nlookupdata nd;
858 	struct thread *td;
859 	struct proc *p;
860 	struct mount *mp;
861 	int error;
862 
863 	td = curthread;
864 	p = td->td_proc;
865 	if (p->p_ucred->cr_prison && !prison_quotas)
866 		return (EPERM);
867 
868 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
869 	if (error == 0)
870 		error = nlookup(&nd);
871 	if (error == 0) {
872 		mp = nd.nl_nch.mount;
873 		error = VFS_QUOTACTL(mp, uap->cmd, uap->uid,
874 				    uap->arg, nd.nl_cred);
875 	}
876 	nlookup_done(&nd);
877 	return (error);
878 }
879 
880 /*
881  * mountctl(char *path, int op, int fd, const void *ctl, int ctllen,
882  *		void *buf, int buflen)
883  *
884  * This function operates on a mount point and executes the specified
885  * operation using the specified control data, and possibly returns data.
886  *
887  * The actual number of bytes stored in the result buffer is returned, 0
888  * if none, otherwise an error is returned.
889  */
890 /* ARGSUSED */
891 int
892 sys_mountctl(struct mountctl_args *uap)
893 {
894 	struct thread *td = curthread;
895 	struct proc *p = td->td_proc;
896 	struct file *fp;
897 	void *ctl = NULL;
898 	void *buf = NULL;
899 	char *path = NULL;
900 	int error;
901 
902 	/*
903 	 * Sanity and permissions checks.  We must be root.
904 	 */
905 	KKASSERT(p);
906 	if (p->p_ucred->cr_prison != NULL)
907 		return (EPERM);
908 	if ((error = suser(td)) != 0)
909 		return (error);
910 
911 	/*
912 	 * Argument length checks
913 	 */
914 	if (uap->ctllen < 0 || uap->ctllen > 1024)
915 		return (EINVAL);
916 	if (uap->buflen < 0 || uap->buflen > 16 * 1024)
917 		return (EINVAL);
918 	if (uap->path == NULL)
919 		return (EINVAL);
920 
921 	/*
922 	 * Allocate the necessary buffers and copyin data
923 	 */
924 	path = objcache_get(namei_oc, M_WAITOK);
925 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
926 	if (error)
927 		goto done;
928 
929 	if (uap->ctllen) {
930 		ctl = kmalloc(uap->ctllen + 1, M_TEMP, M_WAITOK|M_ZERO);
931 		error = copyin(uap->ctl, ctl, uap->ctllen);
932 		if (error)
933 			goto done;
934 	}
935 	if (uap->buflen)
936 		buf = kmalloc(uap->buflen + 1, M_TEMP, M_WAITOK|M_ZERO);
937 
938 	/*
939 	 * Validate the descriptor
940 	 */
941 	if (uap->fd >= 0) {
942 		fp = holdfp(p->p_fd, uap->fd, -1);
943 		if (fp == NULL) {
944 			error = EBADF;
945 			goto done;
946 		}
947 	} else {
948 		fp = NULL;
949 	}
950 
951 	/*
952 	 * Execute the internal kernel function and clean up.
953 	 */
954 	error = kern_mountctl(path, uap->op, fp, ctl, uap->ctllen, buf, uap->buflen, &uap->sysmsg_result);
955 	if (fp)
956 		fdrop(fp);
957 	if (error == 0 && uap->sysmsg_result > 0)
958 		error = copyout(buf, uap->buf, uap->sysmsg_result);
959 done:
960 	if (path)
961 		objcache_put(namei_oc, path);
962 	if (ctl)
963 		kfree(ctl, M_TEMP);
964 	if (buf)
965 		kfree(buf, M_TEMP);
966 	return (error);
967 }
968 
969 /*
970  * Execute a mount control operation by resolving the path to a mount point
971  * and calling vop_mountctl().
972  *
973  * Use the mount point from the nch instead of the vnode so nullfs mounts
974  * can properly spike the VOP.
975  */
976 int
977 kern_mountctl(const char *path, int op, struct file *fp,
978 		const void *ctl, int ctllen,
979 		void *buf, int buflen, int *res)
980 {
981 	struct vnode *vp;
982 	struct mount *mp;
983 	struct nlookupdata nd;
984 	int error;
985 
986 	*res = 0;
987 	vp = NULL;
988 	error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_FOLLOW);
989 	if (error == 0)
990 		error = nlookup(&nd);
991 	if (error == 0)
992 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
993 	mp = nd.nl_nch.mount;
994 	nlookup_done(&nd);
995 	if (error)
996 		return (error);
997 
998 	/*
999 	 * Must be the root of the filesystem
1000 	 */
1001 	if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) {
1002 		vput(vp);
1003 		return (EINVAL);
1004 	}
1005 	error = vop_mountctl(mp->mnt_vn_use_ops, op, fp, ctl, ctllen,
1006 				buf, buflen, res);
1007 	vput(vp);
1008 	return (error);
1009 }
1010 
1011 int
1012 kern_statfs(struct nlookupdata *nd, struct statfs *buf)
1013 {
1014 	struct thread *td = curthread;
1015 	struct proc *p = td->td_proc;
1016 	struct mount *mp;
1017 	struct statfs *sp;
1018 	char *fullpath, *freepath;
1019 	int error;
1020 
1021 	if ((error = nlookup(nd)) != 0)
1022 		return (error);
1023 	mp = nd->nl_nch.mount;
1024 	sp = &mp->mnt_stat;
1025 	if ((error = VFS_STATFS(mp, sp, nd->nl_cred)) != 0)
1026 		return (error);
1027 
1028 	error = mount_path(p, mp, &fullpath, &freepath);
1029 	if (error)
1030 		return(error);
1031 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1032 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1033 	kfree(freepath, M_TEMP);
1034 
1035 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1036 	bcopy(sp, buf, sizeof(*buf));
1037 	/* Only root should have access to the fsid's. */
1038 	if (suser(td))
1039 		buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1040 	return (0);
1041 }
1042 
1043 /*
1044  * statfs_args(char *path, struct statfs *buf)
1045  *
1046  * Get filesystem statistics.
1047  */
1048 int
1049 sys_statfs(struct statfs_args *uap)
1050 {
1051 	struct nlookupdata nd;
1052 	struct statfs buf;
1053 	int error;
1054 
1055 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1056 	if (error == 0)
1057 		error = kern_statfs(&nd, &buf);
1058 	nlookup_done(&nd);
1059 	if (error == 0)
1060 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1061 	return (error);
1062 }
1063 
1064 int
1065 kern_fstatfs(int fd, struct statfs *buf)
1066 {
1067 	struct thread *td = curthread;
1068 	struct proc *p = td->td_proc;
1069 	struct file *fp;
1070 	struct mount *mp;
1071 	struct statfs *sp;
1072 	char *fullpath, *freepath;
1073 	int error;
1074 
1075 	KKASSERT(p);
1076 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1077 		return (error);
1078 	mp = ((struct vnode *)fp->f_data)->v_mount;
1079 	if (mp == NULL) {
1080 		error = EBADF;
1081 		goto done;
1082 	}
1083 	if (fp->f_cred == NULL) {
1084 		error = EINVAL;
1085 		goto done;
1086 	}
1087 	sp = &mp->mnt_stat;
1088 	if ((error = VFS_STATFS(mp, sp, fp->f_cred)) != 0)
1089 		goto done;
1090 
1091 	if ((error = mount_path(p, mp, &fullpath, &freepath)) != 0)
1092 		goto done;
1093 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1094 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1095 	kfree(freepath, M_TEMP);
1096 
1097 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1098 	bcopy(sp, buf, sizeof(*buf));
1099 
1100 	/* Only root should have access to the fsid's. */
1101 	if (suser(td))
1102 		buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1103 	error = 0;
1104 done:
1105 	fdrop(fp);
1106 	return (error);
1107 }
1108 
1109 /*
1110  * fstatfs_args(int fd, struct statfs *buf)
1111  *
1112  * Get filesystem statistics.
1113  */
1114 int
1115 sys_fstatfs(struct fstatfs_args *uap)
1116 {
1117 	struct statfs buf;
1118 	int error;
1119 
1120 	error = kern_fstatfs(uap->fd, &buf);
1121 
1122 	if (error == 0)
1123 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1124 	return (error);
1125 }
1126 
1127 int
1128 kern_statvfs(struct nlookupdata *nd, struct statvfs *buf)
1129 {
1130 	struct mount *mp;
1131 	struct statvfs *sp;
1132 	int error;
1133 
1134 	if ((error = nlookup(nd)) != 0)
1135 		return (error);
1136 	mp = nd->nl_nch.mount;
1137 	sp = &mp->mnt_vstat;
1138 	if ((error = VFS_STATVFS(mp, sp, nd->nl_cred)) != 0)
1139 		return (error);
1140 
1141 	sp->f_flag = 0;
1142 	if (mp->mnt_flag & MNT_RDONLY)
1143 		sp->f_flag |= ST_RDONLY;
1144 	if (mp->mnt_flag & MNT_NOSUID)
1145 		sp->f_flag |= ST_NOSUID;
1146 	bcopy(sp, buf, sizeof(*buf));
1147 	return (0);
1148 }
1149 
1150 /*
1151  * statfs_args(char *path, struct statfs *buf)
1152  *
1153  * Get filesystem statistics.
1154  */
1155 int
1156 sys_statvfs(struct statvfs_args *uap)
1157 {
1158 	struct nlookupdata nd;
1159 	struct statvfs buf;
1160 	int error;
1161 
1162 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1163 	if (error == 0)
1164 		error = kern_statvfs(&nd, &buf);
1165 	nlookup_done(&nd);
1166 	if (error == 0)
1167 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1168 	return (error);
1169 }
1170 
1171 int
1172 kern_fstatvfs(int fd, struct statvfs *buf)
1173 {
1174 	struct thread *td = curthread;
1175 	struct proc *p = td->td_proc;
1176 	struct file *fp;
1177 	struct mount *mp;
1178 	struct statvfs *sp;
1179 	int error;
1180 
1181 	KKASSERT(p);
1182 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1183 		return (error);
1184 	mp = ((struct vnode *)fp->f_data)->v_mount;
1185 	if (mp == NULL) {
1186 		error = EBADF;
1187 		goto done;
1188 	}
1189 	if (fp->f_cred == NULL) {
1190 		error = EINVAL;
1191 		goto done;
1192 	}
1193 	sp = &mp->mnt_vstat;
1194 	if ((error = VFS_STATVFS(mp, sp, fp->f_cred)) != 0)
1195 		goto done;
1196 
1197 	sp->f_flag = 0;
1198 	if (mp->mnt_flag & MNT_RDONLY)
1199 		sp->f_flag |= ST_RDONLY;
1200 	if (mp->mnt_flag & MNT_NOSUID)
1201 		sp->f_flag |= ST_NOSUID;
1202 
1203 	bcopy(sp, buf, sizeof(*buf));
1204 	error = 0;
1205 done:
1206 	fdrop(fp);
1207 	return (error);
1208 }
1209 
1210 /*
1211  * fstatfs_args(int fd, struct statfs *buf)
1212  *
1213  * Get filesystem statistics.
1214  */
1215 int
1216 sys_fstatvfs(struct fstatvfs_args *uap)
1217 {
1218 	struct statvfs buf;
1219 	int error;
1220 
1221 	error = kern_fstatvfs(uap->fd, &buf);
1222 
1223 	if (error == 0)
1224 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1225 	return (error);
1226 }
1227 
1228 /*
1229  * getfsstat_args(struct statfs *buf, long bufsize, int flags)
1230  *
1231  * Get statistics on all filesystems.
1232  */
1233 
1234 struct getfsstat_info {
1235 	struct statfs *sfsp;
1236 	long count;
1237 	long maxcount;
1238 	int error;
1239 	int flags;
1240 	struct proc *p;
1241 };
1242 
1243 static int getfsstat_callback(struct mount *, void *);
1244 
1245 /* ARGSUSED */
1246 int
1247 sys_getfsstat(struct getfsstat_args *uap)
1248 {
1249 	struct thread *td = curthread;
1250 	struct proc *p = td->td_proc;
1251 	struct getfsstat_info info;
1252 
1253 	bzero(&info, sizeof(info));
1254 
1255 	info.maxcount = uap->bufsize / sizeof(struct statfs);
1256 	info.sfsp = uap->buf;
1257 	info.count = 0;
1258 	info.flags = uap->flags;
1259 	info.p = p;
1260 
1261 	mountlist_scan(getfsstat_callback, &info, MNTSCAN_FORWARD);
1262 	if (info.sfsp && info.count > info.maxcount)
1263 		uap->sysmsg_result = info.maxcount;
1264 	else
1265 		uap->sysmsg_result = info.count;
1266 	return (info.error);
1267 }
1268 
1269 static int
1270 getfsstat_callback(struct mount *mp, void *data)
1271 {
1272 	struct getfsstat_info *info = data;
1273 	struct statfs *sp;
1274 	char *freepath;
1275 	char *fullpath;
1276 	int error;
1277 
1278 	if (info->sfsp && info->count < info->maxcount) {
1279 		if (info->p && !chroot_visible_mnt(mp, info->p))
1280 			return(0);
1281 		sp = &mp->mnt_stat;
1282 
1283 		/*
1284 		 * If MNT_NOWAIT or MNT_LAZY is specified, do not
1285 		 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1286 		 * overrides MNT_WAIT.
1287 		 */
1288 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1289 		    (info->flags & MNT_WAIT)) &&
1290 		    (error = VFS_STATFS(mp, sp, info->p->p_ucred))) {
1291 			return(0);
1292 		}
1293 		sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1294 
1295 		error = mount_path(info->p, mp, &fullpath, &freepath);
1296 		if (error) {
1297 			info->error = error;
1298 			return(-1);
1299 		}
1300 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1301 		strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1302 		kfree(freepath, M_TEMP);
1303 
1304 		error = copyout(sp, info->sfsp, sizeof(*sp));
1305 		if (error) {
1306 			info->error = error;
1307 			return (-1);
1308 		}
1309 		++info->sfsp;
1310 	}
1311 	info->count++;
1312 	return(0);
1313 }
1314 
1315 /*
1316  * getvfsstat_args(struct statfs *buf, struct statvfs *vbuf,
1317 		   long bufsize, int flags)
1318  *
1319  * Get statistics on all filesystems.
1320  */
1321 
1322 struct getvfsstat_info {
1323 	struct statfs *sfsp;
1324 	struct statvfs *vsfsp;
1325 	long count;
1326 	long maxcount;
1327 	int error;
1328 	int flags;
1329 	struct proc *p;
1330 };
1331 
1332 static int getvfsstat_callback(struct mount *, void *);
1333 
1334 /* ARGSUSED */
1335 int
1336 sys_getvfsstat(struct getvfsstat_args *uap)
1337 {
1338 	struct thread *td = curthread;
1339 	struct proc *p = td->td_proc;
1340 	struct getvfsstat_info info;
1341 
1342 	bzero(&info, sizeof(info));
1343 
1344 	info.maxcount = uap->vbufsize / sizeof(struct statvfs);
1345 	info.sfsp = uap->buf;
1346 	info.vsfsp = uap->vbuf;
1347 	info.count = 0;
1348 	info.flags = uap->flags;
1349 	info.p = p;
1350 
1351 	mountlist_scan(getvfsstat_callback, &info, MNTSCAN_FORWARD);
1352 	if (info.vsfsp && info.count > info.maxcount)
1353 		uap->sysmsg_result = info.maxcount;
1354 	else
1355 		uap->sysmsg_result = info.count;
1356 	return (info.error);
1357 }
1358 
1359 static int
1360 getvfsstat_callback(struct mount *mp, void *data)
1361 {
1362 	struct getvfsstat_info *info = data;
1363 	struct statfs *sp;
1364 	struct statvfs *vsp;
1365 	char *freepath;
1366 	char *fullpath;
1367 	int error;
1368 
1369 	if (info->vsfsp && info->count < info->maxcount) {
1370 		if (info->p && !chroot_visible_mnt(mp, info->p))
1371 			return(0);
1372 		sp = &mp->mnt_stat;
1373 		vsp = &mp->mnt_vstat;
1374 
1375 		/*
1376 		 * If MNT_NOWAIT or MNT_LAZY is specified, do not
1377 		 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1378 		 * overrides MNT_WAIT.
1379 		 */
1380 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1381 		    (info->flags & MNT_WAIT)) &&
1382 		    (error = VFS_STATFS(mp, sp, info->p->p_ucred))) {
1383 			return(0);
1384 		}
1385 		sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1386 
1387 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1388 		    (info->flags & MNT_WAIT)) &&
1389 		    (error = VFS_STATVFS(mp, vsp, info->p->p_ucred))) {
1390 			return(0);
1391 		}
1392 		vsp->f_flag = 0;
1393 		if (mp->mnt_flag & MNT_RDONLY)
1394 			vsp->f_flag |= ST_RDONLY;
1395 		if (mp->mnt_flag & MNT_NOSUID)
1396 			vsp->f_flag |= ST_NOSUID;
1397 
1398 		error = mount_path(info->p, mp, &fullpath, &freepath);
1399 		if (error) {
1400 			info->error = error;
1401 			return(-1);
1402 		}
1403 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1404 		strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1405 		kfree(freepath, M_TEMP);
1406 
1407 		error = copyout(sp, info->sfsp, sizeof(*sp));
1408 		if (error == 0)
1409 			error = copyout(vsp, info->vsfsp, sizeof(*vsp));
1410 		if (error) {
1411 			info->error = error;
1412 			return (-1);
1413 		}
1414 		++info->sfsp;
1415 		++info->vsfsp;
1416 	}
1417 	info->count++;
1418 	return(0);
1419 }
1420 
1421 
1422 /*
1423  * fchdir_args(int fd)
1424  *
1425  * Change current working directory to a given file descriptor.
1426  */
1427 /* ARGSUSED */
1428 int
1429 sys_fchdir(struct fchdir_args *uap)
1430 {
1431 	struct thread *td = curthread;
1432 	struct proc *p = td->td_proc;
1433 	struct filedesc *fdp = p->p_fd;
1434 	struct vnode *vp, *ovp;
1435 	struct mount *mp;
1436 	struct file *fp;
1437 	struct nchandle nch, onch, tnch;
1438 	int error;
1439 
1440 	if ((error = holdvnode(fdp, uap->fd, &fp)) != 0)
1441 		return (error);
1442 	vp = (struct vnode *)fp->f_data;
1443 	vref(vp);
1444 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1445 	if (vp->v_type != VDIR || fp->f_nchandle.ncp == NULL)
1446 		error = ENOTDIR;
1447 	else
1448 		error = VOP_ACCESS(vp, VEXEC, p->p_ucred);
1449 	if (error) {
1450 		vput(vp);
1451 		fdrop(fp);
1452 		return (error);
1453 	}
1454 	cache_copy(&fp->f_nchandle, &nch);
1455 
1456 	/*
1457 	 * If the ncp has become a mount point, traverse through
1458 	 * the mount point.
1459 	 */
1460 
1461 	while (!error && (nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
1462 	       (mp = cache_findmount(&nch)) != NULL
1463 	) {
1464 		error = nlookup_mp(mp, &tnch);
1465 		if (error == 0) {
1466 			cache_unlock(&tnch);	/* leave ref intact */
1467 			vput(vp);
1468 			vp = tnch.ncp->nc_vp;
1469 			error = vget(vp, LK_SHARED);
1470 			KKASSERT(error == 0);
1471 			cache_drop(&nch);
1472 			nch = tnch;
1473 		}
1474 	}
1475 	if (error == 0) {
1476 		ovp = fdp->fd_cdir;
1477 		onch = fdp->fd_ncdir;
1478 		vn_unlock(vp);		/* leave ref intact */
1479 		fdp->fd_cdir = vp;
1480 		fdp->fd_ncdir = nch;
1481 		cache_drop(&onch);
1482 		vrele(ovp);
1483 	} else {
1484 		cache_drop(&nch);
1485 		vput(vp);
1486 	}
1487 	fdrop(fp);
1488 	return (error);
1489 }
1490 
1491 int
1492 kern_chdir(struct nlookupdata *nd)
1493 {
1494 	struct thread *td = curthread;
1495 	struct proc *p = td->td_proc;
1496 	struct filedesc *fdp = p->p_fd;
1497 	struct vnode *vp, *ovp;
1498 	struct nchandle onch;
1499 	int error;
1500 
1501 	if ((error = nlookup(nd)) != 0)
1502 		return (error);
1503 	if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
1504 		return (ENOENT);
1505 	if ((error = vget(vp, LK_SHARED)) != 0)
1506 		return (error);
1507 
1508 	error = checkvp_chdir(vp, td);
1509 	vn_unlock(vp);
1510 	if (error == 0) {
1511 		ovp = fdp->fd_cdir;
1512 		onch = fdp->fd_ncdir;
1513 		cache_unlock(&nd->nl_nch);	/* leave reference intact */
1514 		fdp->fd_ncdir = nd->nl_nch;
1515 		fdp->fd_cdir = vp;
1516 		cache_drop(&onch);
1517 		vrele(ovp);
1518 		cache_zero(&nd->nl_nch);
1519 	} else {
1520 		vrele(vp);
1521 	}
1522 	return (error);
1523 }
1524 
1525 /*
1526  * chdir_args(char *path)
1527  *
1528  * Change current working directory (``.'').
1529  */
1530 int
1531 sys_chdir(struct chdir_args *uap)
1532 {
1533 	struct nlookupdata nd;
1534 	int error;
1535 
1536 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1537 	if (error == 0)
1538 		error = kern_chdir(&nd);
1539 	nlookup_done(&nd);
1540 	return (error);
1541 }
1542 
1543 /*
1544  * Helper function for raised chroot(2) security function:  Refuse if
1545  * any filedescriptors are open directories.
1546  */
1547 static int
1548 chroot_refuse_vdir_fds(struct filedesc *fdp)
1549 {
1550 	struct vnode *vp;
1551 	struct file *fp;
1552 	int error;
1553 	int fd;
1554 
1555 	for (fd = 0; fd < fdp->fd_nfiles ; fd++) {
1556 		if ((error = holdvnode(fdp, fd, &fp)) != 0)
1557 			continue;
1558 		vp = (struct vnode *)fp->f_data;
1559 		if (vp->v_type != VDIR) {
1560 			fdrop(fp);
1561 			continue;
1562 		}
1563 		fdrop(fp);
1564 		return(EPERM);
1565 	}
1566 	return (0);
1567 }
1568 
1569 /*
1570  * This sysctl determines if we will allow a process to chroot(2) if it
1571  * has a directory open:
1572  *	0: disallowed for all processes.
1573  *	1: allowed for processes that were not already chroot(2)'ed.
1574  *	2: allowed for all processes.
1575  */
1576 
1577 static int chroot_allow_open_directories = 1;
1578 
1579 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
1580      &chroot_allow_open_directories, 0, "");
1581 
1582 /*
1583  * chroot to the specified namecache entry.  We obtain the vp from the
1584  * namecache data.  The passed ncp must be locked and referenced and will
1585  * remain locked and referenced on return.
1586  */
1587 int
1588 kern_chroot(struct nchandle *nch)
1589 {
1590 	struct thread *td = curthread;
1591 	struct proc *p = td->td_proc;
1592 	struct filedesc *fdp = p->p_fd;
1593 	struct vnode *vp;
1594 	int error;
1595 
1596 	/*
1597 	 * Only root can chroot
1598 	 */
1599 	if ((error = suser_cred(p->p_ucred, PRISON_ROOT)) != 0)
1600 		return (error);
1601 
1602 	/*
1603 	 * Disallow open directory descriptors (fchdir() breakouts).
1604 	 */
1605 	if (chroot_allow_open_directories == 0 ||
1606 	   (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
1607 		if ((error = chroot_refuse_vdir_fds(fdp)) != 0)
1608 			return (error);
1609 	}
1610 	if ((vp = nch->ncp->nc_vp) == NULL)
1611 		return (ENOENT);
1612 
1613 	if ((error = vget(vp, LK_SHARED)) != 0)
1614 		return (error);
1615 
1616 	/*
1617 	 * Check the validity of vp as a directory to change to and
1618 	 * associate it with rdir/jdir.
1619 	 */
1620 	error = checkvp_chdir(vp, td);
1621 	vn_unlock(vp);			/* leave reference intact */
1622 	if (error == 0) {
1623 		vrele(fdp->fd_rdir);
1624 		fdp->fd_rdir = vp;	/* reference inherited by fd_rdir */
1625 		cache_drop(&fdp->fd_nrdir);
1626 		cache_copy(nch, &fdp->fd_nrdir);
1627 		if (fdp->fd_jdir == NULL) {
1628 			fdp->fd_jdir = vp;
1629 			vref(fdp->fd_jdir);
1630 			cache_copy(nch, &fdp->fd_njdir);
1631 		}
1632 	} else {
1633 		vrele(vp);
1634 	}
1635 	return (error);
1636 }
1637 
1638 /*
1639  * chroot_args(char *path)
1640  *
1641  * Change notion of root (``/'') directory.
1642  */
1643 /* ARGSUSED */
1644 int
1645 sys_chroot(struct chroot_args *uap)
1646 {
1647 	struct thread *td = curthread;
1648 	struct nlookupdata nd;
1649 	int error;
1650 
1651 	KKASSERT(td->td_proc);
1652 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1653 	if (error) {
1654 		nlookup_done(&nd);
1655 		return(error);
1656 	}
1657 	error = nlookup(&nd);
1658 	if (error == 0)
1659 		error = kern_chroot(&nd.nl_nch);
1660 	nlookup_done(&nd);
1661 	return(error);
1662 }
1663 
1664 /*
1665  * Common routine for chroot and chdir.  Given a locked, referenced vnode,
1666  * determine whether it is legal to chdir to the vnode.  The vnode's state
1667  * is not changed by this call.
1668  */
1669 int
1670 checkvp_chdir(struct vnode *vp, struct thread *td)
1671 {
1672 	int error;
1673 
1674 	if (vp->v_type != VDIR)
1675 		error = ENOTDIR;
1676 	else
1677 		error = VOP_ACCESS(vp, VEXEC, td->td_proc->p_ucred);
1678 	return (error);
1679 }
1680 
1681 int
1682 kern_open(struct nlookupdata *nd, int oflags, int mode, int *res)
1683 {
1684 	struct thread *td = curthread;
1685 	struct proc *p = td->td_proc;
1686 	struct lwp *lp = td->td_lwp;
1687 	struct filedesc *fdp = p->p_fd;
1688 	int cmode, flags;
1689 	struct file *nfp;
1690 	struct file *fp;
1691 	struct vnode *vp;
1692 	int type, indx, error;
1693 	struct flock lf;
1694 
1695 	if ((oflags & O_ACCMODE) == O_ACCMODE)
1696 		return (EINVAL);
1697 	flags = FFLAGS(oflags);
1698 	error = falloc(p, &nfp, NULL);
1699 	if (error)
1700 		return (error);
1701 	fp = nfp;
1702 	cmode = ((mode &~ fdp->fd_cmask) & ALLPERMS) &~ S_ISTXT;
1703 
1704 	/*
1705 	 * XXX p_dupfd is a real mess.  It allows a device to return a
1706 	 * file descriptor to be duplicated rather then doing the open
1707 	 * itself.
1708 	 */
1709 	lp->lwp_dupfd = -1;
1710 
1711 	/*
1712 	 * Call vn_open() to do the lookup and assign the vnode to the
1713 	 * file pointer.  vn_open() does not change the ref count on fp
1714 	 * and the vnode, on success, will be inherited by the file pointer
1715 	 * and unlocked.
1716 	 */
1717 	nd->nl_flags |= NLC_LOCKVP;
1718 	error = vn_open(nd, fp, flags, cmode);
1719 	nlookup_done(nd);
1720 	if (error) {
1721 		/*
1722 		 * handle special fdopen() case.  bleh.  dupfdopen() is
1723 		 * responsible for dropping the old contents of ofiles[indx]
1724 		 * if it succeeds.
1725 		 *
1726 		 * Note that fsetfd() will add a ref to fp which represents
1727 		 * the fd_files[] assignment.  We must still drop our
1728 		 * reference.
1729 		 */
1730 		if ((error == ENODEV || error == ENXIO) && lp->lwp_dupfd >= 0) {
1731 			if (fdalloc(p, 0, &indx) == 0) {
1732 				error = dupfdopen(p, indx, lp->lwp_dupfd, flags, error);
1733 				if (error == 0) {
1734 					*res = indx;
1735 					fdrop(fp);	/* our ref */
1736 					return (0);
1737 				}
1738 				fsetfd(p, NULL, indx);
1739 			}
1740 		}
1741 		fdrop(fp);	/* our ref */
1742 		if (error == ERESTART)
1743 			error = EINTR;
1744 		return (error);
1745 	}
1746 
1747 	/*
1748 	 * ref the vnode for ourselves so it can't be ripped out from under
1749 	 * is.  XXX need an ND flag to request that the vnode be returned
1750 	 * anyway.
1751 	 *
1752 	 * Reserve a file descriptor but do not assign it until the open
1753 	 * succeeds.
1754 	 */
1755 	vp = (struct vnode *)fp->f_data;
1756 	vref(vp);
1757 	if ((error = fdalloc(p, 0, &indx)) != 0) {
1758 		fdrop(fp);
1759 		vrele(vp);
1760 		return (error);
1761 	}
1762 
1763 	/*
1764 	 * If no error occurs the vp will have been assigned to the file
1765 	 * pointer.
1766 	 */
1767 	lp->lwp_dupfd = 0;
1768 
1769 	if (flags & (O_EXLOCK | O_SHLOCK)) {
1770 		lf.l_whence = SEEK_SET;
1771 		lf.l_start = 0;
1772 		lf.l_len = 0;
1773 		if (flags & O_EXLOCK)
1774 			lf.l_type = F_WRLCK;
1775 		else
1776 			lf.l_type = F_RDLCK;
1777 		if (flags & FNONBLOCK)
1778 			type = 0;
1779 		else
1780 			type = F_WAIT;
1781 
1782 		if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
1783 			/*
1784 			 * lock request failed.  Clean up the reserved
1785 			 * descriptor.
1786 			 */
1787 			vrele(vp);
1788 			fsetfd(p, NULL, indx);
1789 			fdrop(fp);
1790 			return (error);
1791 		}
1792 		fp->f_flag |= FHASLOCK;
1793 	}
1794 #if 0
1795 	/*
1796 	 * Assert that all regular file vnodes were created with a object.
1797 	 */
1798 	KASSERT(vp->v_type != VREG || vp->v_object != NULL,
1799 		("open: regular file has no backing object after vn_open"));
1800 #endif
1801 
1802 	vrele(vp);
1803 
1804 	/*
1805 	 * release our private reference, leaving the one associated with the
1806 	 * descriptor table intact.
1807 	 */
1808 	fsetfd(p, fp, indx);
1809 	fdrop(fp);
1810 	*res = indx;
1811 	return (0);
1812 }
1813 
1814 /*
1815  * open_args(char *path, int flags, int mode)
1816  *
1817  * Check permissions, allocate an open file structure,
1818  * and call the device open routine if any.
1819  */
1820 int
1821 sys_open(struct open_args *uap)
1822 {
1823 	struct nlookupdata nd;
1824 	int error;
1825 
1826 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1827 	if (error == 0) {
1828 		error = kern_open(&nd, uap->flags,
1829 				    uap->mode, &uap->sysmsg_result);
1830 	}
1831 	nlookup_done(&nd);
1832 	return (error);
1833 }
1834 
1835 int
1836 kern_mknod(struct nlookupdata *nd, int mode, int rmajor, int rminor)
1837 {
1838 	struct thread *td = curthread;
1839 	struct proc *p = td->td_proc;
1840 	struct vnode *vp;
1841 	struct vattr vattr;
1842 	int error;
1843 	int whiteout = 0;
1844 
1845 	KKASSERT(p);
1846 
1847 	switch (mode & S_IFMT) {
1848 	case S_IFCHR:
1849 	case S_IFBLK:
1850 		error = suser(td);
1851 		break;
1852 	default:
1853 		error = suser_cred(p->p_ucred, PRISON_ROOT);
1854 		break;
1855 	}
1856 	if (error)
1857 		return (error);
1858 
1859 	bwillinode(1);
1860 	nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
1861 	if ((error = nlookup(nd)) != 0)
1862 		return (error);
1863 	if (nd->nl_nch.ncp->nc_vp)
1864 		return (EEXIST);
1865 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1866 		return (error);
1867 
1868 	VATTR_NULL(&vattr);
1869 	vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1870 	vattr.va_rmajor = rmajor;
1871 	vattr.va_rminor = rminor;
1872 	whiteout = 0;
1873 
1874 	switch (mode & S_IFMT) {
1875 	case S_IFMT:	/* used by badsect to flag bad sectors */
1876 		vattr.va_type = VBAD;
1877 		break;
1878 	case S_IFCHR:
1879 		vattr.va_type = VCHR;
1880 		break;
1881 	case S_IFBLK:
1882 		vattr.va_type = VBLK;
1883 		break;
1884 	case S_IFWHT:
1885 		whiteout = 1;
1886 		break;
1887 	case S_IFDIR:
1888 		/* special directories support for HAMMER */
1889 		vattr.va_type = VDIR;
1890 		break;
1891 	default:
1892 		error = EINVAL;
1893 		break;
1894 	}
1895 	if (error == 0) {
1896 		if (whiteout) {
1897 			error = VOP_NWHITEOUT(&nd->nl_nch, nd->nl_dvp,
1898 					      nd->nl_cred, NAMEI_CREATE);
1899 		} else {
1900 			vp = NULL;
1901 			error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp,
1902 					   &vp, nd->nl_cred, &vattr);
1903 			if (error == 0)
1904 				vput(vp);
1905 		}
1906 	}
1907 	return (error);
1908 }
1909 
1910 /*
1911  * mknod_args(char *path, int mode, int dev)
1912  *
1913  * Create a special file.
1914  */
1915 int
1916 sys_mknod(struct mknod_args *uap)
1917 {
1918 	struct nlookupdata nd;
1919 	int error;
1920 
1921 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1922 	if (error == 0) {
1923 		error = kern_mknod(&nd, uap->mode,
1924 				   umajor(uap->dev), uminor(uap->dev));
1925 	}
1926 	nlookup_done(&nd);
1927 	return (error);
1928 }
1929 
1930 int
1931 kern_mkfifo(struct nlookupdata *nd, int mode)
1932 {
1933 	struct thread *td = curthread;
1934 	struct proc *p = td->td_proc;
1935 	struct vattr vattr;
1936 	struct vnode *vp;
1937 	int error;
1938 
1939 	bwillinode(1);
1940 
1941 	nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
1942 	if ((error = nlookup(nd)) != 0)
1943 		return (error);
1944 	if (nd->nl_nch.ncp->nc_vp)
1945 		return (EEXIST);
1946 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1947 		return (error);
1948 
1949 	VATTR_NULL(&vattr);
1950 	vattr.va_type = VFIFO;
1951 	vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1952 	vp = NULL;
1953 	error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp, &vp, nd->nl_cred, &vattr);
1954 	if (error == 0)
1955 		vput(vp);
1956 	return (error);
1957 }
1958 
1959 /*
1960  * mkfifo_args(char *path, int mode)
1961  *
1962  * Create a named pipe.
1963  */
1964 int
1965 sys_mkfifo(struct mkfifo_args *uap)
1966 {
1967 	struct nlookupdata nd;
1968 	int error;
1969 
1970 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1971 	if (error == 0)
1972 		error = kern_mkfifo(&nd, uap->mode);
1973 	nlookup_done(&nd);
1974 	return (error);
1975 }
1976 
1977 static int hardlink_check_uid = 0;
1978 SYSCTL_INT(_security, OID_AUTO, hardlink_check_uid, CTLFLAG_RW,
1979     &hardlink_check_uid, 0,
1980     "Unprivileged processes cannot create hard links to files owned by other "
1981     "users");
1982 static int hardlink_check_gid = 0;
1983 SYSCTL_INT(_security, OID_AUTO, hardlink_check_gid, CTLFLAG_RW,
1984     &hardlink_check_gid, 0,
1985     "Unprivileged processes cannot create hard links to files owned by other "
1986     "groups");
1987 
1988 static int
1989 can_hardlink(struct vnode *vp, struct thread *td, struct ucred *cred)
1990 {
1991 	struct vattr va;
1992 	int error;
1993 
1994 	/*
1995 	 * Shortcut if disabled
1996 	 */
1997 	if (hardlink_check_uid == 0 && hardlink_check_gid == 0)
1998 		return (0);
1999 
2000 	/*
2001 	 * root cred can always hardlink
2002 	 */
2003 	if (suser_cred(cred, PRISON_ROOT) == 0)
2004 		return (0);
2005 
2006 	/*
2007 	 * Otherwise only if the originating file is owned by the
2008 	 * same user or group.  Note that any group is allowed if
2009 	 * the file is owned by the caller.
2010 	 */
2011 	error = VOP_GETATTR(vp, &va);
2012 	if (error != 0)
2013 		return (error);
2014 
2015 	if (hardlink_check_uid) {
2016 		if (cred->cr_uid != va.va_uid)
2017 			return (EPERM);
2018 	}
2019 
2020 	if (hardlink_check_gid) {
2021 		if (cred->cr_uid != va.va_uid && !groupmember(va.va_gid, cred))
2022 			return (EPERM);
2023 	}
2024 
2025 	return (0);
2026 }
2027 
2028 int
2029 kern_link(struct nlookupdata *nd, struct nlookupdata *linknd)
2030 {
2031 	struct thread *td = curthread;
2032 	struct vnode *vp;
2033 	int error;
2034 
2035 	/*
2036 	 * Lookup the source and obtained a locked vnode.
2037 	 *
2038 	 * XXX relookup on vget failure / race ?
2039 	 */
2040 	bwillinode(1);
2041 	if ((error = nlookup(nd)) != 0)
2042 		return (error);
2043 	vp = nd->nl_nch.ncp->nc_vp;
2044 	KKASSERT(vp != NULL);
2045 	if (vp->v_type == VDIR)
2046 		return (EPERM);		/* POSIX */
2047 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2048 		return (error);
2049 	if ((error = vget(vp, LK_EXCLUSIVE)) != 0)
2050 		return (error);
2051 
2052 	/*
2053 	 * Unlock the source so we can lookup the target without deadlocking
2054 	 * (XXX vp is locked already, possible other deadlock?).  The target
2055 	 * must not exist.
2056 	 */
2057 	KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
2058 	nd->nl_flags &= ~NLC_NCPISLOCKED;
2059 	cache_unlock(&nd->nl_nch);
2060 
2061 	linknd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2062 	if ((error = nlookup(linknd)) != 0) {
2063 		vput(vp);
2064 		return (error);
2065 	}
2066 	if (linknd->nl_nch.ncp->nc_vp) {
2067 		vput(vp);
2068 		return (EEXIST);
2069 	}
2070 
2071 	/*
2072 	 * Finally run the new API VOP.
2073 	 */
2074 	error = can_hardlink(vp, td, td->td_proc->p_ucred);
2075 	if (error == 0) {
2076 		error = VOP_NLINK(&linknd->nl_nch, linknd->nl_dvp,
2077 				  vp, linknd->nl_cred);
2078 	}
2079 	vput(vp);
2080 	return (error);
2081 }
2082 
2083 /*
2084  * link_args(char *path, char *link)
2085  *
2086  * Make a hard file link.
2087  */
2088 int
2089 sys_link(struct link_args *uap)
2090 {
2091 	struct nlookupdata nd, linknd;
2092 	int error;
2093 
2094 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2095 	if (error == 0) {
2096 		error = nlookup_init(&linknd, uap->link, UIO_USERSPACE, 0);
2097 		if (error == 0)
2098 			error = kern_link(&nd, &linknd);
2099 		nlookup_done(&linknd);
2100 	}
2101 	nlookup_done(&nd);
2102 	return (error);
2103 }
2104 
2105 int
2106 kern_symlink(struct nlookupdata *nd, char *path, int mode)
2107 {
2108 	struct vattr vattr;
2109 	struct vnode *vp;
2110 	struct vnode *dvp;
2111 	int error;
2112 
2113 	bwillinode(1);
2114 	nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2115 	if ((error = nlookup(nd)) != 0)
2116 		return (error);
2117 	if (nd->nl_nch.ncp->nc_vp)
2118 		return (EEXIST);
2119 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2120 		return (error);
2121 	dvp = nd->nl_dvp;
2122 	VATTR_NULL(&vattr);
2123 	vattr.va_mode = mode;
2124 	error = VOP_NSYMLINK(&nd->nl_nch, dvp, &vp, nd->nl_cred, &vattr, path);
2125 	if (error == 0)
2126 		vput(vp);
2127 	return (error);
2128 }
2129 
2130 /*
2131  * symlink(char *path, char *link)
2132  *
2133  * Make a symbolic link.
2134  */
2135 int
2136 sys_symlink(struct symlink_args *uap)
2137 {
2138 	struct thread *td = curthread;
2139 	struct nlookupdata nd;
2140 	char *path;
2141 	int error;
2142 	int mode;
2143 
2144 	path = objcache_get(namei_oc, M_WAITOK);
2145 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
2146 	if (error == 0) {
2147 		error = nlookup_init(&nd, uap->link, UIO_USERSPACE, 0);
2148 		if (error == 0) {
2149 			mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
2150 			error = kern_symlink(&nd, path, mode);
2151 		}
2152 		nlookup_done(&nd);
2153 	}
2154 	objcache_put(namei_oc, path);
2155 	return (error);
2156 }
2157 
2158 /*
2159  * undelete_args(char *path)
2160  *
2161  * Delete a whiteout from the filesystem.
2162  */
2163 /* ARGSUSED */
2164 int
2165 sys_undelete(struct undelete_args *uap)
2166 {
2167 	struct nlookupdata nd;
2168 	int error;
2169 
2170 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2171 	bwillinode(1);
2172 	nd.nl_flags |= NLC_DELETE | NLC_REFDVP;
2173 	if (error == 0)
2174 		error = nlookup(&nd);
2175 	if (error == 0)
2176 		error = ncp_writechk(&nd.nl_nch);
2177 	if (error == 0) {
2178 		error = VOP_NWHITEOUT(&nd.nl_nch, nd.nl_dvp, nd.nl_cred,
2179 				      NAMEI_DELETE);
2180 	}
2181 	nlookup_done(&nd);
2182 	return (error);
2183 }
2184 
2185 int
2186 kern_unlink(struct nlookupdata *nd)
2187 {
2188 	int error;
2189 
2190 	bwillinode(1);
2191 	nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
2192 	if ((error = nlookup(nd)) != 0)
2193 		return (error);
2194 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2195 		return (error);
2196 	error = VOP_NREMOVE(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
2197 	return (error);
2198 }
2199 
2200 /*
2201  * unlink_args(char *path)
2202  *
2203  * Delete a name from the filesystem.
2204  */
2205 int
2206 sys_unlink(struct unlink_args *uap)
2207 {
2208 	struct nlookupdata nd;
2209 	int error;
2210 
2211 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2212 	if (error == 0)
2213 		error = kern_unlink(&nd);
2214 	nlookup_done(&nd);
2215 	return (error);
2216 }
2217 
2218 int
2219 kern_lseek(int fd, off_t offset, int whence, off_t *res)
2220 {
2221 	struct thread *td = curthread;
2222 	struct proc *p = td->td_proc;
2223 	struct file *fp;
2224 	struct vnode *vp;
2225 	struct vattr vattr;
2226 	off_t new_offset;
2227 	int error;
2228 
2229 	fp = holdfp(p->p_fd, fd, -1);
2230 	if (fp == NULL)
2231 		return (EBADF);
2232 	if (fp->f_type != DTYPE_VNODE) {
2233 		error = ESPIPE;
2234 		goto done;
2235 	}
2236 	vp = (struct vnode *)fp->f_data;
2237 
2238 	switch (whence) {
2239 	case L_INCR:
2240 		new_offset = fp->f_offset + offset;
2241 		error = 0;
2242 		break;
2243 	case L_XTND:
2244 		error = VOP_GETATTR(vp, &vattr);
2245 		new_offset = offset + vattr.va_size;
2246 		break;
2247 	case L_SET:
2248 		new_offset = offset;
2249 		error = 0;
2250 		break;
2251 	default:
2252 		new_offset = 0;
2253 		error = EINVAL;
2254 		break;
2255 	}
2256 
2257 	/*
2258 	 * Validate the seek position.  Negative offsets are not allowed
2259 	 * for regular files, block specials, or directories.
2260 	 */
2261 	if (error == 0) {
2262 		if (new_offset < 0 &&
2263 		    (vp->v_type == VREG || vp->v_type == VDIR ||
2264 		     vp->v_type == VCHR || vp->v_type == VBLK)) {
2265 			error = EINVAL;
2266 		} else {
2267 			fp->f_offset = new_offset;
2268 		}
2269 	}
2270 	*res = fp->f_offset;
2271 done:
2272 	fdrop(fp);
2273 	return (error);
2274 }
2275 
2276 /*
2277  * lseek_args(int fd, int pad, off_t offset, int whence)
2278  *
2279  * Reposition read/write file offset.
2280  */
2281 int
2282 sys_lseek(struct lseek_args *uap)
2283 {
2284 	int error;
2285 
2286 	error = kern_lseek(uap->fd, uap->offset, uap->whence,
2287 			   &uap->sysmsg_offset);
2288 
2289 	return (error);
2290 }
2291 
2292 int
2293 kern_access(struct nlookupdata *nd, int aflags)
2294 {
2295 	struct vnode *vp;
2296 	int error, flags;
2297 
2298 	if ((error = nlookup(nd)) != 0)
2299 		return (error);
2300 retry:
2301 	error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2302 	if (error)
2303 		return (error);
2304 
2305 	/* Flags == 0 means only check for existence. */
2306 	if (aflags) {
2307 		flags = 0;
2308 		if (aflags & R_OK)
2309 			flags |= VREAD;
2310 		if (aflags & W_OK)
2311 			flags |= VWRITE;
2312 		if (aflags & X_OK)
2313 			flags |= VEXEC;
2314 		if ((flags & VWRITE) == 0 ||
2315 		    (error = vn_writechk(vp, &nd->nl_nch)) == 0)
2316 			error = VOP_ACCESS(vp, flags, nd->nl_cred);
2317 
2318 		/*
2319 		 * If the file handle is stale we have to re-resolve the
2320 		 * entry.  This is a hack at the moment.
2321 		 */
2322 		if (error == ESTALE) {
2323 			vput(vp);
2324 			cache_setunresolved(&nd->nl_nch);
2325 			error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2326 			if (error == 0) {
2327 				vp = NULL;
2328 				goto retry;
2329 			}
2330 			return(error);
2331 		}
2332 	}
2333 	vput(vp);
2334 	return (error);
2335 }
2336 
2337 /*
2338  * access_args(char *path, int flags)
2339  *
2340  * Check access permissions.
2341  */
2342 int
2343 sys_access(struct access_args *uap)
2344 {
2345 	struct nlookupdata nd;
2346 	int error;
2347 
2348 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2349 	if (error == 0)
2350 		error = kern_access(&nd, uap->flags);
2351 	nlookup_done(&nd);
2352 	return (error);
2353 }
2354 
2355 int
2356 kern_stat(struct nlookupdata *nd, struct stat *st)
2357 {
2358 	int error;
2359 	struct vnode *vp;
2360 	thread_t td;
2361 
2362 	if ((error = nlookup(nd)) != 0)
2363 		return (error);
2364 again:
2365 	if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
2366 		return (ENOENT);
2367 
2368 	td = curthread;
2369 	if ((error = vget(vp, LK_SHARED)) != 0)
2370 		return (error);
2371 	error = vn_stat(vp, st, nd->nl_cred);
2372 
2373 	/*
2374 	 * If the file handle is stale we have to re-resolve the entry.  This
2375 	 * is a hack at the moment.
2376 	 */
2377 	if (error == ESTALE) {
2378 		vput(vp);
2379 		cache_setunresolved(&nd->nl_nch);
2380 		error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2381 		if (error == 0)
2382 			goto again;
2383 	} else {
2384 		vput(vp);
2385 	}
2386 	return (error);
2387 }
2388 
2389 /*
2390  * stat_args(char *path, struct stat *ub)
2391  *
2392  * Get file status; this version follows links.
2393  */
2394 int
2395 sys_stat(struct stat_args *uap)
2396 {
2397 	struct nlookupdata nd;
2398 	struct stat st;
2399 	int error;
2400 
2401 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2402 	if (error == 0) {
2403 		error = kern_stat(&nd, &st);
2404 		if (error == 0)
2405 			error = copyout(&st, uap->ub, sizeof(*uap->ub));
2406 	}
2407 	nlookup_done(&nd);
2408 	return (error);
2409 }
2410 
2411 /*
2412  * lstat_args(char *path, struct stat *ub)
2413  *
2414  * Get file status; this version does not follow links.
2415  */
2416 int
2417 sys_lstat(struct lstat_args *uap)
2418 {
2419 	struct nlookupdata nd;
2420 	struct stat st;
2421 	int error;
2422 
2423 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2424 	if (error == 0) {
2425 		error = kern_stat(&nd, &st);
2426 		if (error == 0)
2427 			error = copyout(&st, uap->ub, sizeof(*uap->ub));
2428 	}
2429 	nlookup_done(&nd);
2430 	return (error);
2431 }
2432 
2433 /*
2434  * pathconf_Args(char *path, int name)
2435  *
2436  * Get configurable pathname variables.
2437  */
2438 /* ARGSUSED */
2439 int
2440 sys_pathconf(struct pathconf_args *uap)
2441 {
2442 	struct nlookupdata nd;
2443 	struct vnode *vp;
2444 	int error;
2445 
2446 	vp = NULL;
2447 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2448 	if (error == 0)
2449 		error = nlookup(&nd);
2450 	if (error == 0)
2451 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
2452 	nlookup_done(&nd);
2453 	if (error == 0) {
2454 		error = VOP_PATHCONF(vp, uap->name, uap->sysmsg_fds);
2455 		vput(vp);
2456 	}
2457 	return (error);
2458 }
2459 
2460 /*
2461  * XXX: daver
2462  * kern_readlink isn't properly split yet.  There is a copyin burried
2463  * in VOP_READLINK().
2464  */
2465 int
2466 kern_readlink(struct nlookupdata *nd, char *buf, int count, int *res)
2467 {
2468 	struct thread *td = curthread;
2469 	struct proc *p = td->td_proc;
2470 	struct vnode *vp;
2471 	struct iovec aiov;
2472 	struct uio auio;
2473 	int error;
2474 
2475 	if ((error = nlookup(nd)) != 0)
2476 		return (error);
2477 	error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2478 	if (error)
2479 		return (error);
2480 	if (vp->v_type != VLNK) {
2481 		error = EINVAL;
2482 	} else {
2483 		aiov.iov_base = buf;
2484 		aiov.iov_len = count;
2485 		auio.uio_iov = &aiov;
2486 		auio.uio_iovcnt = 1;
2487 		auio.uio_offset = 0;
2488 		auio.uio_rw = UIO_READ;
2489 		auio.uio_segflg = UIO_USERSPACE;
2490 		auio.uio_td = td;
2491 		auio.uio_resid = count;
2492 		error = VOP_READLINK(vp, &auio, p->p_ucred);
2493 	}
2494 	vput(vp);
2495 	*res = count - auio.uio_resid;
2496 	return (error);
2497 }
2498 
2499 /*
2500  * readlink_args(char *path, char *buf, int count)
2501  *
2502  * Return target name of a symbolic link.
2503  */
2504 int
2505 sys_readlink(struct readlink_args *uap)
2506 {
2507 	struct nlookupdata nd;
2508 	int error;
2509 
2510 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2511 	if (error == 0) {
2512 		error = kern_readlink(&nd, uap->buf, uap->count,
2513 					&uap->sysmsg_result);
2514 	}
2515 	nlookup_done(&nd);
2516 	return (error);
2517 }
2518 
2519 static int
2520 setfflags(struct vnode *vp, int flags)
2521 {
2522 	struct thread *td = curthread;
2523 	struct proc *p = td->td_proc;
2524 	int error;
2525 	struct vattr vattr;
2526 
2527 	/*
2528 	 * Prevent non-root users from setting flags on devices.  When
2529 	 * a device is reused, users can retain ownership of the device
2530 	 * if they are allowed to set flags and programs assume that
2531 	 * chown can't fail when done as root.
2532 	 */
2533 	if ((vp->v_type == VCHR || vp->v_type == VBLK) &&
2534 	    ((error = suser_cred(p->p_ucred, PRISON_ROOT)) != 0))
2535 		return (error);
2536 
2537 	/*
2538 	 * note: vget is required for any operation that might mod the vnode
2539 	 * so VINACTIVE is properly cleared.
2540 	 */
2541 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2542 		VATTR_NULL(&vattr);
2543 		vattr.va_flags = flags;
2544 		error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2545 		vput(vp);
2546 	}
2547 	return (error);
2548 }
2549 
2550 /*
2551  * chflags(char *path, int flags)
2552  *
2553  * Change flags of a file given a path name.
2554  */
2555 /* ARGSUSED */
2556 int
2557 sys_chflags(struct chflags_args *uap)
2558 {
2559 	struct nlookupdata nd;
2560 	struct vnode *vp;
2561 	int error;
2562 
2563 	vp = NULL;
2564 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2565 	/* XXX Add NLC flag indicating modifying operation? */
2566 	if (error == 0)
2567 		error = nlookup(&nd);
2568 	if (error == 0)
2569 		error = ncp_writechk(&nd.nl_nch);
2570 	if (error == 0)
2571 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
2572 	nlookup_done(&nd);
2573 	if (error == 0) {
2574 		error = setfflags(vp, uap->flags);
2575 		vrele(vp);
2576 	}
2577 	return (error);
2578 }
2579 
2580 /*
2581  * lchflags(char *path, int flags)
2582  *
2583  * Change flags of a file given a path name, but don't follow symlinks.
2584  */
2585 /* ARGSUSED */
2586 int
2587 sys_lchflags(struct lchflags_args *uap)
2588 {
2589 	struct nlookupdata nd;
2590 	struct vnode *vp;
2591 	int error;
2592 
2593 	vp = NULL;
2594 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2595 	/* XXX Add NLC flag indicating modifying operation? */
2596 	if (error == 0)
2597 		error = nlookup(&nd);
2598 	if (error == 0)
2599 		error = ncp_writechk(&nd.nl_nch);
2600 	if (error == 0)
2601 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
2602 	nlookup_done(&nd);
2603 	if (error == 0) {
2604 		error = setfflags(vp, uap->flags);
2605 		vrele(vp);
2606 	}
2607 	return (error);
2608 }
2609 
2610 /*
2611  * fchflags_args(int fd, int flags)
2612  *
2613  * Change flags of a file given a file descriptor.
2614  */
2615 /* ARGSUSED */
2616 int
2617 sys_fchflags(struct fchflags_args *uap)
2618 {
2619 	struct thread *td = curthread;
2620 	struct proc *p = td->td_proc;
2621 	struct file *fp;
2622 	int error;
2623 
2624 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2625 		return (error);
2626 	if (fp->f_nchandle.ncp)
2627 		error = ncp_writechk(&fp->f_nchandle);
2628 	if (error == 0)
2629 		error = setfflags((struct vnode *) fp->f_data, uap->flags);
2630 	fdrop(fp);
2631 	return (error);
2632 }
2633 
2634 static int
2635 setfmode(struct vnode *vp, int mode)
2636 {
2637 	struct thread *td = curthread;
2638 	struct proc *p = td->td_proc;
2639 	int error;
2640 	struct vattr vattr;
2641 
2642 	/*
2643 	 * note: vget is required for any operation that might mod the vnode
2644 	 * so VINACTIVE is properly cleared.
2645 	 */
2646 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2647 		VATTR_NULL(&vattr);
2648 		vattr.va_mode = mode & ALLPERMS;
2649 		error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2650 		vput(vp);
2651 	}
2652 	return error;
2653 }
2654 
2655 int
2656 kern_chmod(struct nlookupdata *nd, int mode)
2657 {
2658 	struct vnode *vp;
2659 	int error;
2660 
2661 	/* XXX Add NLC flag indicating modifying operation? */
2662 	if ((error = nlookup(nd)) != 0)
2663 		return (error);
2664 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2665 		return (error);
2666 	if ((error = ncp_writechk(&nd->nl_nch)) == 0)
2667 		error = setfmode(vp, mode);
2668 	vrele(vp);
2669 	return (error);
2670 }
2671 
2672 /*
2673  * chmod_args(char *path, int mode)
2674  *
2675  * Change mode of a file given path name.
2676  */
2677 /* ARGSUSED */
2678 int
2679 sys_chmod(struct chmod_args *uap)
2680 {
2681 	struct nlookupdata nd;
2682 	int error;
2683 
2684 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2685 	if (error == 0)
2686 		error = kern_chmod(&nd, uap->mode);
2687 	nlookup_done(&nd);
2688 	return (error);
2689 }
2690 
2691 /*
2692  * lchmod_args(char *path, int mode)
2693  *
2694  * Change mode of a file given path name (don't follow links.)
2695  */
2696 /* ARGSUSED */
2697 int
2698 sys_lchmod(struct lchmod_args *uap)
2699 {
2700 	struct nlookupdata nd;
2701 	int error;
2702 
2703 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2704 	if (error == 0)
2705 		error = kern_chmod(&nd, uap->mode);
2706 	nlookup_done(&nd);
2707 	return (error);
2708 }
2709 
2710 /*
2711  * fchmod_args(int fd, int mode)
2712  *
2713  * Change mode of a file given a file descriptor.
2714  */
2715 /* ARGSUSED */
2716 int
2717 sys_fchmod(struct fchmod_args *uap)
2718 {
2719 	struct thread *td = curthread;
2720 	struct proc *p = td->td_proc;
2721 	struct file *fp;
2722 	int error;
2723 
2724 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2725 		return (error);
2726 	if (fp->f_nchandle.ncp)
2727 		error = ncp_writechk(&fp->f_nchandle);
2728 	if (error == 0)
2729 		error = setfmode((struct vnode *)fp->f_data, uap->mode);
2730 	fdrop(fp);
2731 	return (error);
2732 }
2733 
2734 static int
2735 setfown(struct vnode *vp, uid_t uid, gid_t gid)
2736 {
2737 	struct thread *td = curthread;
2738 	struct proc *p = td->td_proc;
2739 	int error;
2740 	struct vattr vattr;
2741 
2742 	/*
2743 	 * note: vget is required for any operation that might mod the vnode
2744 	 * so VINACTIVE is properly cleared.
2745 	 */
2746 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2747 		VATTR_NULL(&vattr);
2748 		vattr.va_uid = uid;
2749 		vattr.va_gid = gid;
2750 		error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2751 		vput(vp);
2752 	}
2753 	return error;
2754 }
2755 
2756 int
2757 kern_chown(struct nlookupdata *nd, int uid, int gid)
2758 {
2759 	struct vnode *vp;
2760 	int error;
2761 
2762 	/* XXX Add NLC flag indicating modifying operation? */
2763 	if ((error = nlookup(nd)) != 0)
2764 		return (error);
2765 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2766 		return (error);
2767 	if ((error = ncp_writechk(&nd->nl_nch)) == 0)
2768 		error = setfown(vp, uid, gid);
2769 	vrele(vp);
2770 	return (error);
2771 }
2772 
2773 /*
2774  * chown(char *path, int uid, int gid)
2775  *
2776  * Set ownership given a path name.
2777  */
2778 int
2779 sys_chown(struct chown_args *uap)
2780 {
2781 	struct nlookupdata nd;
2782 	int error;
2783 
2784 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2785 	if (error == 0)
2786 		error = kern_chown(&nd, uap->uid, uap->gid);
2787 	nlookup_done(&nd);
2788 	return (error);
2789 }
2790 
2791 /*
2792  * lchown_args(char *path, int uid, int gid)
2793  *
2794  * Set ownership given a path name, do not cross symlinks.
2795  */
2796 int
2797 sys_lchown(struct lchown_args *uap)
2798 {
2799 	struct nlookupdata nd;
2800 	int error;
2801 
2802 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2803 	if (error == 0)
2804 		error = kern_chown(&nd, uap->uid, uap->gid);
2805 	nlookup_done(&nd);
2806 	return (error);
2807 }
2808 
2809 /*
2810  * fchown_args(int fd, int uid, int gid)
2811  *
2812  * Set ownership given a file descriptor.
2813  */
2814 /* ARGSUSED */
2815 int
2816 sys_fchown(struct fchown_args *uap)
2817 {
2818 	struct thread *td = curthread;
2819 	struct proc *p = td->td_proc;
2820 	struct file *fp;
2821 	int error;
2822 
2823 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2824 		return (error);
2825 	if (fp->f_nchandle.ncp)
2826 		error = ncp_writechk(&fp->f_nchandle);
2827 	if (error == 0)
2828 		error = setfown((struct vnode *)fp->f_data, uap->uid, uap->gid);
2829 	fdrop(fp);
2830 	return (error);
2831 }
2832 
2833 static int
2834 getutimes(const struct timeval *tvp, struct timespec *tsp)
2835 {
2836 	struct timeval tv[2];
2837 
2838 	if (tvp == NULL) {
2839 		microtime(&tv[0]);
2840 		TIMEVAL_TO_TIMESPEC(&tv[0], &tsp[0]);
2841 		tsp[1] = tsp[0];
2842 	} else {
2843 		TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]);
2844 		TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]);
2845 	}
2846 	return 0;
2847 }
2848 
2849 static int
2850 setutimes(struct vnode *vp, const struct timespec *ts, int nullflag)
2851 {
2852 	struct thread *td = curthread;
2853 	struct proc *p = td->td_proc;
2854 	int error;
2855 	struct vattr vattr;
2856 
2857 	/*
2858 	 * note: vget is required for any operation that might mod the vnode
2859 	 * so VINACTIVE is properly cleared.
2860 	 */
2861 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2862 		VATTR_NULL(&vattr);
2863 		vattr.va_atime = ts[0];
2864 		vattr.va_mtime = ts[1];
2865 		if (nullflag)
2866 			vattr.va_vaflags |= VA_UTIMES_NULL;
2867 		error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2868 		vput(vp);
2869 	}
2870 	return error;
2871 }
2872 
2873 int
2874 kern_utimes(struct nlookupdata *nd, struct timeval *tptr)
2875 {
2876 	struct timespec ts[2];
2877 	struct vnode *vp;
2878 	int error;
2879 
2880 	if ((error = getutimes(tptr, ts)) != 0)
2881 		return (error);
2882 	/* XXX Add NLC flag indicating modifying operation? */
2883 	if ((error = nlookup(nd)) != 0)
2884 		return (error);
2885 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2886 		return (error);
2887 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2888 		return (error);
2889 	error = setutimes(vp, ts, tptr == NULL);
2890 	vrele(vp);
2891 	return (error);
2892 }
2893 
2894 /*
2895  * utimes_args(char *path, struct timeval *tptr)
2896  *
2897  * Set the access and modification times of a file.
2898  */
2899 int
2900 sys_utimes(struct utimes_args *uap)
2901 {
2902 	struct timeval tv[2];
2903 	struct nlookupdata nd;
2904 	int error;
2905 
2906 	if (uap->tptr) {
2907  		error = copyin(uap->tptr, tv, sizeof(tv));
2908 		if (error)
2909 			return (error);
2910 	}
2911 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2912 	if (error == 0)
2913 		error = kern_utimes(&nd, uap->tptr ? tv : NULL);
2914 	nlookup_done(&nd);
2915 	return (error);
2916 }
2917 
2918 /*
2919  * lutimes_args(char *path, struct timeval *tptr)
2920  *
2921  * Set the access and modification times of a file.
2922  */
2923 int
2924 sys_lutimes(struct lutimes_args *uap)
2925 {
2926 	struct timeval tv[2];
2927 	struct nlookupdata nd;
2928 	int error;
2929 
2930 	if (uap->tptr) {
2931 		error = copyin(uap->tptr, tv, sizeof(tv));
2932 		if (error)
2933 			return (error);
2934 	}
2935 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2936 	if (error == 0)
2937 		error = kern_utimes(&nd, uap->tptr ? tv : NULL);
2938 	nlookup_done(&nd);
2939 	return (error);
2940 }
2941 
2942 int
2943 kern_futimes(int fd, struct timeval *tptr)
2944 {
2945 	struct thread *td = curthread;
2946 	struct proc *p = td->td_proc;
2947 	struct timespec ts[2];
2948 	struct file *fp;
2949 	int error;
2950 
2951 	error = getutimes(tptr, ts);
2952 	if (error)
2953 		return (error);
2954 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
2955 		return (error);
2956 	if (fp->f_nchandle.ncp)
2957 		error = ncp_writechk(&fp->f_nchandle);
2958 	if (error == 0)
2959 		error =  setutimes((struct vnode *)fp->f_data, ts, tptr == NULL);
2960 	fdrop(fp);
2961 	return (error);
2962 }
2963 
2964 /*
2965  * futimes_args(int fd, struct timeval *tptr)
2966  *
2967  * Set the access and modification times of a file.
2968  */
2969 int
2970 sys_futimes(struct futimes_args *uap)
2971 {
2972 	struct timeval tv[2];
2973 	int error;
2974 
2975 	if (uap->tptr) {
2976 		error = copyin(uap->tptr, tv, sizeof(tv));
2977 		if (error)
2978 			return (error);
2979 	}
2980 
2981 	error = kern_futimes(uap->fd, uap->tptr ? tv : NULL);
2982 
2983 	return (error);
2984 }
2985 
2986 int
2987 kern_truncate(struct nlookupdata *nd, off_t length)
2988 {
2989 	struct vnode *vp;
2990 	struct vattr vattr;
2991 	int error;
2992 
2993 	if (length < 0)
2994 		return(EINVAL);
2995 	/* XXX Add NLC flag indicating modifying operation? */
2996 	if ((error = nlookup(nd)) != 0)
2997 		return (error);
2998 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2999 		return (error);
3000 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3001 		return (error);
3002 	if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) {
3003 		vrele(vp);
3004 		return (error);
3005 	}
3006 	if (vp->v_type == VDIR) {
3007 		error = EISDIR;
3008 	} else if ((error = vn_writechk(vp, &nd->nl_nch)) == 0 &&
3009 	    (error = VOP_ACCESS(vp, VWRITE, nd->nl_cred)) == 0) {
3010 		VATTR_NULL(&vattr);
3011 		vattr.va_size = length;
3012 		error = VOP_SETATTR(vp, &vattr, nd->nl_cred);
3013 	}
3014 	vput(vp);
3015 	return (error);
3016 }
3017 
3018 /*
3019  * truncate(char *path, int pad, off_t length)
3020  *
3021  * Truncate a file given its path name.
3022  */
3023 int
3024 sys_truncate(struct truncate_args *uap)
3025 {
3026 	struct nlookupdata nd;
3027 	int error;
3028 
3029 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3030 	if (error == 0)
3031 		error = kern_truncate(&nd, uap->length);
3032 	nlookup_done(&nd);
3033 	return error;
3034 }
3035 
3036 int
3037 kern_ftruncate(int fd, off_t length)
3038 {
3039 	struct thread *td = curthread;
3040 	struct proc *p = td->td_proc;
3041 	struct vattr vattr;
3042 	struct vnode *vp;
3043 	struct file *fp;
3044 	int error;
3045 
3046 	if (length < 0)
3047 		return(EINVAL);
3048 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3049 		return (error);
3050 	if (fp->f_nchandle.ncp) {
3051 		error = ncp_writechk(&fp->f_nchandle);
3052 		if (error)
3053 			goto done;
3054 	}
3055 	if ((fp->f_flag & FWRITE) == 0) {
3056 		error = EINVAL;
3057 		goto done;
3058 	}
3059 	vp = (struct vnode *)fp->f_data;
3060 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3061 	if (vp->v_type == VDIR) {
3062 		error = EISDIR;
3063 	} else if ((error = vn_writechk(vp, NULL)) == 0) {
3064 		VATTR_NULL(&vattr);
3065 		vattr.va_size = length;
3066 		error = VOP_SETATTR(vp, &vattr, fp->f_cred);
3067 	}
3068 	vn_unlock(vp);
3069 done:
3070 	fdrop(fp);
3071 	return (error);
3072 }
3073 
3074 /*
3075  * ftruncate_args(int fd, int pad, off_t length)
3076  *
3077  * Truncate a file given a file descriptor.
3078  */
3079 int
3080 sys_ftruncate(struct ftruncate_args *uap)
3081 {
3082 	int error;
3083 
3084 	error = kern_ftruncate(uap->fd, uap->length);
3085 
3086 	return (error);
3087 }
3088 
3089 /*
3090  * fsync(int fd)
3091  *
3092  * Sync an open file.
3093  */
3094 /* ARGSUSED */
3095 int
3096 sys_fsync(struct fsync_args *uap)
3097 {
3098 	struct thread *td = curthread;
3099 	struct proc *p = td->td_proc;
3100 	struct vnode *vp;
3101 	struct file *fp;
3102 	vm_object_t obj;
3103 	int error;
3104 
3105 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3106 		return (error);
3107 	vp = (struct vnode *)fp->f_data;
3108 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3109 	if ((obj = vp->v_object) != NULL)
3110 		vm_object_page_clean(obj, 0, 0, 0);
3111 	if ((error = VOP_FSYNC(vp, MNT_WAIT)) == 0 && vp->v_mount)
3112 		error = buf_fsync(vp);
3113 	vn_unlock(vp);
3114 	fdrop(fp);
3115 	return (error);
3116 }
3117 
3118 int
3119 kern_rename(struct nlookupdata *fromnd, struct nlookupdata *tond)
3120 {
3121 	struct nchandle fnchd;
3122 	struct nchandle tnchd;
3123 	struct namecache *ncp;
3124 	struct vnode *fdvp;
3125 	struct vnode *tdvp;
3126 	struct mount *mp;
3127 	int error;
3128 
3129 	bwillinode(1);
3130 	fromnd->nl_flags |= NLC_REFDVP;
3131 	if ((error = nlookup(fromnd)) != 0)
3132 		return (error);
3133 	if ((fnchd.ncp = fromnd->nl_nch.ncp->nc_parent) == NULL)
3134 		return (ENOENT);
3135 	fnchd.mount = fromnd->nl_nch.mount;
3136 	cache_hold(&fnchd);
3137 
3138 	/*
3139 	 * unlock the source nch so we can lookup the target nch without
3140 	 * deadlocking.  The target may or may not exist so we do not check
3141 	 * for a target vp like kern_mkdir() and other creation functions do.
3142 	 *
3143 	 * The source and target directories are ref'd and rechecked after
3144 	 * everything is relocked to determine if the source or target file
3145 	 * has been renamed.
3146 	 */
3147 	KKASSERT(fromnd->nl_flags & NLC_NCPISLOCKED);
3148 	fromnd->nl_flags &= ~NLC_NCPISLOCKED;
3149 	cache_unlock(&fromnd->nl_nch);
3150 
3151 	tond->nl_flags |= NLC_CREATE | NLC_REFDVP;
3152 	if ((error = nlookup(tond)) != 0) {
3153 		cache_drop(&fnchd);
3154 		return (error);
3155 	}
3156 	if ((tnchd.ncp = tond->nl_nch.ncp->nc_parent) == NULL) {
3157 		cache_drop(&fnchd);
3158 		return (ENOENT);
3159 	}
3160 	tnchd.mount = tond->nl_nch.mount;
3161 	cache_hold(&tnchd);
3162 
3163 	/*
3164 	 * If the source and target are the same there is nothing to do
3165 	 */
3166 	if (fromnd->nl_nch.ncp == tond->nl_nch.ncp) {
3167 		cache_drop(&fnchd);
3168 		cache_drop(&tnchd);
3169 		return (0);
3170 	}
3171 
3172 	/*
3173 	 * Mount points cannot be renamed or overwritten
3174 	 */
3175 	if ((fromnd->nl_nch.ncp->nc_flag | tond->nl_nch.ncp->nc_flag) &
3176 	    NCF_ISMOUNTPT
3177 	) {
3178 		cache_drop(&fnchd);
3179 		cache_drop(&tnchd);
3180 		return (EINVAL);
3181 	}
3182 
3183 	/*
3184 	 * relock the source ncp.  NOTE AFTER RELOCKING: the source ncp
3185 	 * may have become invalid while it was unlocked, nc_vp and nc_mount
3186 	 * could be NULL.
3187 	 */
3188 	if (cache_lock_nonblock(&fromnd->nl_nch) == 0) {
3189 		cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
3190 	} else if (fromnd->nl_nch.ncp > tond->nl_nch.ncp) {
3191 		cache_lock(&fromnd->nl_nch);
3192 		cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
3193 	} else {
3194 		cache_unlock(&tond->nl_nch);
3195 		cache_lock(&fromnd->nl_nch);
3196 		cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
3197 		cache_lock(&tond->nl_nch);
3198 		cache_resolve(&tond->nl_nch, tond->nl_cred);
3199 	}
3200 	fromnd->nl_flags |= NLC_NCPISLOCKED;
3201 
3202 	/*
3203 	 * make sure the parent directories linkages are the same
3204 	 */
3205 	if (fnchd.ncp != fromnd->nl_nch.ncp->nc_parent ||
3206 	    tnchd.ncp != tond->nl_nch.ncp->nc_parent) {
3207 		cache_drop(&fnchd);
3208 		cache_drop(&tnchd);
3209 		return (ENOENT);
3210 	}
3211 
3212 	/*
3213 	 * Both the source and target must be within the same filesystem and
3214 	 * in the same filesystem as their parent directories within the
3215 	 * namecache topology.
3216 	 *
3217 	 * NOTE: fromnd's nc_mount or nc_vp could be NULL.
3218 	 */
3219 	mp = fnchd.mount;
3220 	if (mp != tnchd.mount || mp != fromnd->nl_nch.mount ||
3221 	    mp != tond->nl_nch.mount) {
3222 		cache_drop(&fnchd);
3223 		cache_drop(&tnchd);
3224 		return (EXDEV);
3225 	}
3226 
3227 	/*
3228 	 * Make sure the mount point is writable
3229 	 */
3230 	if ((error = ncp_writechk(&tond->nl_nch)) != 0) {
3231 		cache_drop(&fnchd);
3232 		cache_drop(&tnchd);
3233 		return (error);
3234 	}
3235 
3236 	/*
3237 	 * If the target exists and either the source or target is a directory,
3238 	 * then both must be directories.
3239 	 *
3240 	 * Due to relocking of the source, fromnd->nl_nch.ncp->nc_vp might h
3241 	 * have become NULL.
3242 	 */
3243 	if (tond->nl_nch.ncp->nc_vp) {
3244 		if (fromnd->nl_nch.ncp->nc_vp == NULL) {
3245 			error = ENOENT;
3246 		} else if (fromnd->nl_nch.ncp->nc_vp->v_type == VDIR) {
3247 			if (tond->nl_nch.ncp->nc_vp->v_type != VDIR)
3248 				error = ENOTDIR;
3249 		} else if (tond->nl_nch.ncp->nc_vp->v_type == VDIR) {
3250 			error = EISDIR;
3251 		}
3252 	}
3253 
3254 	/*
3255 	 * You cannot rename a source into itself or a subdirectory of itself.
3256 	 * We check this by travsersing the target directory upwards looking
3257 	 * for a match against the source.
3258 	 */
3259 	if (error == 0) {
3260 		for (ncp = tnchd.ncp; ncp; ncp = ncp->nc_parent) {
3261 			if (fromnd->nl_nch.ncp == ncp) {
3262 				error = EINVAL;
3263 				break;
3264 			}
3265 		}
3266 	}
3267 
3268 	cache_drop(&fnchd);
3269 	cache_drop(&tnchd);
3270 
3271 	/*
3272 	 * Even though the namespaces are different, they may still represent
3273 	 * hardlinks to the same file.  The filesystem might have a hard time
3274 	 * with this so we issue a NREMOVE of the source instead of a NRENAME
3275 	 * when we detect the situation.
3276 	 */
3277 	if (error == 0) {
3278 		fdvp = fromnd->nl_dvp;
3279 		tdvp = tond->nl_dvp;
3280 		if (fdvp == NULL || tdvp == NULL) {
3281 			error = EPERM;
3282 		} else if (fromnd->nl_nch.ncp->nc_vp == tond->nl_nch.ncp->nc_vp) {
3283 			error = VOP_NREMOVE(&fromnd->nl_nch, fdvp,
3284 					    fromnd->nl_cred);
3285 		} else {
3286 			error = VOP_NRENAME(&fromnd->nl_nch, &tond->nl_nch,
3287 					    fdvp, tdvp, tond->nl_cred);
3288 		}
3289 	}
3290 	return (error);
3291 }
3292 
3293 /*
3294  * rename_args(char *from, char *to)
3295  *
3296  * Rename files.  Source and destination must either both be directories,
3297  * or both not be directories.  If target is a directory, it must be empty.
3298  */
3299 int
3300 sys_rename(struct rename_args *uap)
3301 {
3302 	struct nlookupdata fromnd, tond;
3303 	int error;
3304 
3305 	error = nlookup_init(&fromnd, uap->from, UIO_USERSPACE, 0);
3306 	if (error == 0) {
3307 		error = nlookup_init(&tond, uap->to, UIO_USERSPACE, 0);
3308 		if (error == 0)
3309 			error = kern_rename(&fromnd, &tond);
3310 		nlookup_done(&tond);
3311 	}
3312 	nlookup_done(&fromnd);
3313 	return (error);
3314 }
3315 
3316 int
3317 kern_mkdir(struct nlookupdata *nd, int mode)
3318 {
3319 	struct thread *td = curthread;
3320 	struct proc *p = td->td_proc;
3321 	struct vnode *vp;
3322 	struct vattr vattr;
3323 	int error;
3324 
3325 	bwillinode(1);
3326 	nd->nl_flags |= NLC_WILLBEDIR | NLC_CREATE | NLC_REFDVP;
3327 	if ((error = nlookup(nd)) != 0)
3328 		return (error);
3329 
3330 	if (nd->nl_nch.ncp->nc_vp)
3331 		return (EEXIST);
3332 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3333 		return (error);
3334 	VATTR_NULL(&vattr);
3335 	vattr.va_type = VDIR;
3336 	vattr.va_mode = (mode & ACCESSPERMS) &~ p->p_fd->fd_cmask;
3337 
3338 	vp = NULL;
3339 	error = VOP_NMKDIR(&nd->nl_nch, nd->nl_dvp, &vp, p->p_ucred, &vattr);
3340 	if (error == 0)
3341 		vput(vp);
3342 	return (error);
3343 }
3344 
3345 /*
3346  * mkdir_args(char *path, int mode)
3347  *
3348  * Make a directory file.
3349  */
3350 /* ARGSUSED */
3351 int
3352 sys_mkdir(struct mkdir_args *uap)
3353 {
3354 	struct nlookupdata nd;
3355 	int error;
3356 
3357 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3358 	if (error == 0)
3359 		error = kern_mkdir(&nd, uap->mode);
3360 	nlookup_done(&nd);
3361 	return (error);
3362 }
3363 
3364 int
3365 kern_rmdir(struct nlookupdata *nd)
3366 {
3367 	int error;
3368 
3369 	bwillinode(1);
3370 	nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
3371 	if ((error = nlookup(nd)) != 0)
3372 		return (error);
3373 
3374 	/*
3375 	 * Do not allow directories representing mount points to be
3376 	 * deleted, even if empty.  Check write perms on mount point
3377 	 * in case the vnode is aliased (aka nullfs).
3378 	 */
3379 	if (nd->nl_nch.ncp->nc_flag & (NCF_ISMOUNTPT))
3380 		return (EINVAL);
3381 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3382 		return (error);
3383 	error = VOP_NRMDIR(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
3384 	return (error);
3385 }
3386 
3387 /*
3388  * rmdir_args(char *path)
3389  *
3390  * Remove a directory file.
3391  */
3392 /* ARGSUSED */
3393 int
3394 sys_rmdir(struct rmdir_args *uap)
3395 {
3396 	struct nlookupdata nd;
3397 	int error;
3398 
3399 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3400 	if (error == 0)
3401 		error = kern_rmdir(&nd);
3402 	nlookup_done(&nd);
3403 	return (error);
3404 }
3405 
3406 int
3407 kern_getdirentries(int fd, char *buf, u_int count, long *basep, int *res,
3408     enum uio_seg direction)
3409 {
3410 	struct thread *td = curthread;
3411 	struct proc *p = td->td_proc;
3412 	struct vnode *vp;
3413 	struct file *fp;
3414 	struct uio auio;
3415 	struct iovec aiov;
3416 	off_t loff;
3417 	int error, eofflag;
3418 
3419 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3420 		return (error);
3421 	if ((fp->f_flag & FREAD) == 0) {
3422 		error = EBADF;
3423 		goto done;
3424 	}
3425 	vp = (struct vnode *)fp->f_data;
3426 unionread:
3427 	if (vp->v_type != VDIR) {
3428 		error = EINVAL;
3429 		goto done;
3430 	}
3431 	aiov.iov_base = buf;
3432 	aiov.iov_len = count;
3433 	auio.uio_iov = &aiov;
3434 	auio.uio_iovcnt = 1;
3435 	auio.uio_rw = UIO_READ;
3436 	auio.uio_segflg = direction;
3437 	auio.uio_td = td;
3438 	auio.uio_resid = count;
3439 	loff = auio.uio_offset = fp->f_offset;
3440 	error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL, NULL);
3441 	fp->f_offset = auio.uio_offset;
3442 	if (error)
3443 		goto done;
3444 	if (count == auio.uio_resid) {
3445 		if (union_dircheckp) {
3446 			error = union_dircheckp(td, &vp, fp);
3447 			if (error == -1)
3448 				goto unionread;
3449 			if (error)
3450 				goto done;
3451 		}
3452 #if 0
3453 		if ((vp->v_flag & VROOT) &&
3454 		    (vp->v_mount->mnt_flag & MNT_UNION)) {
3455 			struct vnode *tvp = vp;
3456 			vp = vp->v_mount->mnt_vnodecovered;
3457 			vref(vp);
3458 			fp->f_data = vp;
3459 			fp->f_offset = 0;
3460 			vrele(tvp);
3461 			goto unionread;
3462 		}
3463 #endif
3464 	}
3465 
3466 	/*
3467 	 * WARNING!  *basep may not be wide enough to accomodate the
3468 	 * seek offset.   XXX should we hack this to return the upper 32 bits
3469 	 * for offsets greater then 4G?
3470 	 */
3471 	if (basep) {
3472 		*basep = (long)loff;
3473 	}
3474 	*res = count - auio.uio_resid;
3475 done:
3476 	fdrop(fp);
3477 	return (error);
3478 }
3479 
3480 /*
3481  * getdirentries_args(int fd, char *buf, u_int conut, long *basep)
3482  *
3483  * Read a block of directory entries in a file system independent format.
3484  */
3485 int
3486 sys_getdirentries(struct getdirentries_args *uap)
3487 {
3488 	long base;
3489 	int error;
3490 
3491 	error = kern_getdirentries(uap->fd, uap->buf, uap->count, &base,
3492 				   &uap->sysmsg_result, UIO_USERSPACE);
3493 
3494 	if (error == 0 && uap->basep)
3495 		error = copyout(&base, uap->basep, sizeof(*uap->basep));
3496 	return (error);
3497 }
3498 
3499 /*
3500  * getdents_args(int fd, char *buf, size_t count)
3501  */
3502 int
3503 sys_getdents(struct getdents_args *uap)
3504 {
3505 	int error;
3506 
3507 	error = kern_getdirentries(uap->fd, uap->buf, uap->count, NULL,
3508 	    &uap->sysmsg_result, UIO_USERSPACE);
3509 
3510 	return (error);
3511 }
3512 
3513 /*
3514  * umask(int newmask)
3515  *
3516  * Set the mode mask for creation of filesystem nodes.
3517  *
3518  * MP SAFE
3519  */
3520 int
3521 sys_umask(struct umask_args *uap)
3522 {
3523 	struct thread *td = curthread;
3524 	struct proc *p = td->td_proc;
3525 	struct filedesc *fdp;
3526 
3527 	fdp = p->p_fd;
3528 	uap->sysmsg_result = fdp->fd_cmask;
3529 	fdp->fd_cmask = uap->newmask & ALLPERMS;
3530 	return (0);
3531 }
3532 
3533 /*
3534  * revoke(char *path)
3535  *
3536  * Void all references to file by ripping underlying filesystem
3537  * away from vnode.
3538  */
3539 /* ARGSUSED */
3540 int
3541 sys_revoke(struct revoke_args *uap)
3542 {
3543 	struct nlookupdata nd;
3544 	struct vattr vattr;
3545 	struct vnode *vp;
3546 	struct ucred *cred;
3547 	int error;
3548 
3549 	vp = NULL;
3550 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3551 	if (error == 0)
3552 		error = nlookup(&nd);
3553 	if (error == 0)
3554 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
3555 	cred = crhold(nd.nl_cred);
3556 	nlookup_done(&nd);
3557 	if (error == 0) {
3558 		if (vp->v_type != VCHR && vp->v_type != VBLK)
3559 			error = EINVAL;
3560 		if (error == 0)
3561 			error = VOP_GETATTR(vp, &vattr);
3562 		if (error == 0 && cred->cr_uid != vattr.va_uid)
3563 			error = suser_cred(cred, PRISON_ROOT);
3564 		if (error == 0 && count_udev(vp->v_umajor, vp->v_uminor) > 0) {
3565 			error = 0;
3566 			vx_lock(vp);
3567 			VOP_REVOKE(vp, REVOKEALL);
3568 			vx_unlock(vp);
3569 		}
3570 		vrele(vp);
3571 	}
3572 	if (cred)
3573 		crfree(cred);
3574 	return (error);
3575 }
3576 
3577 /*
3578  * getfh_args(char *fname, fhandle_t *fhp)
3579  *
3580  * Get (NFS) file handle
3581  *
3582  * NOTE: We use the fsid of the covering mount, even if it is a nullfs
3583  * mount.  This allows nullfs mounts to be explicitly exported.
3584  *
3585  * WARNING: nullfs mounts of HAMMER PFS ROOTs are safe.
3586  *
3587  * 	    nullfs mounts of subdirectories are not safe.  That is, it will
3588  *	    work, but you do not really have protection against access to
3589  *	    the related parent directories.
3590  */
3591 int
3592 sys_getfh(struct getfh_args *uap)
3593 {
3594 	struct thread *td = curthread;
3595 	struct nlookupdata nd;
3596 	fhandle_t fh;
3597 	struct vnode *vp;
3598 	struct mount *mp;
3599 	int error;
3600 
3601 	/*
3602 	 * Must be super user
3603 	 */
3604 	if ((error = suser(td)) != 0)
3605 		return (error);
3606 
3607 	vp = NULL;
3608 	error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
3609 	if (error == 0)
3610 		error = nlookup(&nd);
3611 	if (error == 0)
3612 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3613 	mp = nd.nl_nch.mount;
3614 	nlookup_done(&nd);
3615 	if (error == 0) {
3616 		bzero(&fh, sizeof(fh));
3617 		fh.fh_fsid = mp->mnt_stat.f_fsid;
3618 		error = VFS_VPTOFH(vp, &fh.fh_fid);
3619 		vput(vp);
3620 		if (error == 0)
3621 			error = copyout(&fh, uap->fhp, sizeof(fh));
3622 	}
3623 	return (error);
3624 }
3625 
3626 /*
3627  * fhopen_args(const struct fhandle *u_fhp, int flags)
3628  *
3629  * syscall for the rpc.lockd to use to translate a NFS file handle into
3630  * an open descriptor.
3631  *
3632  * warning: do not remove the suser() call or this becomes one giant
3633  * security hole.
3634  */
3635 int
3636 sys_fhopen(struct fhopen_args *uap)
3637 {
3638 	struct thread *td = curthread;
3639 	struct proc *p = td->td_proc;
3640 	struct mount *mp;
3641 	struct vnode *vp;
3642 	struct fhandle fhp;
3643 	struct vattr vat;
3644 	struct vattr *vap = &vat;
3645 	struct flock lf;
3646 	int fmode, mode, error, type;
3647 	struct file *nfp;
3648 	struct file *fp;
3649 	int indx;
3650 
3651 	/*
3652 	 * Must be super user
3653 	 */
3654 	error = suser(td);
3655 	if (error)
3656 		return (error);
3657 
3658 	fmode = FFLAGS(uap->flags);
3659 	/* why not allow a non-read/write open for our lockd? */
3660 	if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
3661 		return (EINVAL);
3662 	error = copyin(uap->u_fhp, &fhp, sizeof(fhp));
3663 	if (error)
3664 		return(error);
3665 	/* find the mount point */
3666 	mp = vfs_getvfs(&fhp.fh_fsid);
3667 	if (mp == NULL)
3668 		return (ESTALE);
3669 	/* now give me my vnode, it gets returned to me locked */
3670 	error = VFS_FHTOVP(mp, NULL, &fhp.fh_fid, &vp);
3671 	if (error)
3672 		return (error);
3673  	/*
3674 	 * from now on we have to make sure not
3675 	 * to forget about the vnode
3676 	 * any error that causes an abort must vput(vp)
3677 	 * just set error = err and 'goto bad;'.
3678 	 */
3679 
3680 	/*
3681 	 * from vn_open
3682 	 */
3683 	if (vp->v_type == VLNK) {
3684 		error = EMLINK;
3685 		goto bad;
3686 	}
3687 	if (vp->v_type == VSOCK) {
3688 		error = EOPNOTSUPP;
3689 		goto bad;
3690 	}
3691 	mode = 0;
3692 	if (fmode & (FWRITE | O_TRUNC)) {
3693 		if (vp->v_type == VDIR) {
3694 			error = EISDIR;
3695 			goto bad;
3696 		}
3697 		error = vn_writechk(vp, NULL);
3698 		if (error)
3699 			goto bad;
3700 		mode |= VWRITE;
3701 	}
3702 	if (fmode & FREAD)
3703 		mode |= VREAD;
3704 	if (mode) {
3705 		error = VOP_ACCESS(vp, mode, p->p_ucred);
3706 		if (error)
3707 			goto bad;
3708 	}
3709 	if (fmode & O_TRUNC) {
3710 		vn_unlock(vp);				/* XXX */
3711 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);	/* XXX */
3712 		VATTR_NULL(vap);
3713 		vap->va_size = 0;
3714 		error = VOP_SETATTR(vp, vap, p->p_ucred);
3715 		if (error)
3716 			goto bad;
3717 	}
3718 
3719 	/*
3720 	 * VOP_OPEN needs the file pointer so it can potentially override
3721 	 * it.
3722 	 *
3723 	 * WARNING! no f_nchandle will be associated when fhopen()ing a
3724 	 * directory.  XXX
3725 	 */
3726 	if ((error = falloc(p, &nfp, &indx)) != 0)
3727 		goto bad;
3728 	fp = nfp;
3729 
3730 	error = VOP_OPEN(vp, fmode, p->p_ucred, fp);
3731 	if (error) {
3732 		/*
3733 		 * setting f_ops this way prevents VOP_CLOSE from being
3734 		 * called or fdrop() releasing the vp from v_data.   Since
3735 		 * the VOP_OPEN failed we don't want to VOP_CLOSE.
3736 		 */
3737 		fp->f_ops = &badfileops;
3738 		fp->f_data = NULL;
3739 		goto bad_drop;
3740 	}
3741 
3742 	/*
3743 	 * The fp is given its own reference, we still have our ref and lock.
3744 	 *
3745 	 * Assert that all regular files must be created with a VM object.
3746 	 */
3747 	if (vp->v_type == VREG && vp->v_object == NULL) {
3748 		kprintf("fhopen: regular file did not have VM object: %p\n", vp);
3749 		goto bad_drop;
3750 	}
3751 
3752 	/*
3753 	 * The open was successful.  Handle any locking requirements.
3754 	 */
3755 	if (fmode & (O_EXLOCK | O_SHLOCK)) {
3756 		lf.l_whence = SEEK_SET;
3757 		lf.l_start = 0;
3758 		lf.l_len = 0;
3759 		if (fmode & O_EXLOCK)
3760 			lf.l_type = F_WRLCK;
3761 		else
3762 			lf.l_type = F_RDLCK;
3763 		if (fmode & FNONBLOCK)
3764 			type = 0;
3765 		else
3766 			type = F_WAIT;
3767 		vn_unlock(vp);
3768 		if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
3769 			/*
3770 			 * release our private reference.
3771 			 */
3772 			fsetfd(p, NULL, indx);
3773 			fdrop(fp);
3774 			vrele(vp);
3775 			return (error);
3776 		}
3777 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3778 		fp->f_flag |= FHASLOCK;
3779 	}
3780 
3781 	/*
3782 	 * Clean up.  Associate the file pointer with the previously
3783 	 * reserved descriptor and return it.
3784 	 */
3785 	vput(vp);
3786 	fsetfd(p, fp, indx);
3787 	fdrop(fp);
3788 	uap->sysmsg_result = indx;
3789 	return (0);
3790 
3791 bad_drop:
3792 	fsetfd(p, NULL, indx);
3793 	fdrop(fp);
3794 bad:
3795 	vput(vp);
3796 	return (error);
3797 }
3798 
3799 /*
3800  * fhstat_args(struct fhandle *u_fhp, struct stat *sb)
3801  */
3802 int
3803 sys_fhstat(struct fhstat_args *uap)
3804 {
3805 	struct thread *td = curthread;
3806 	struct stat sb;
3807 	fhandle_t fh;
3808 	struct mount *mp;
3809 	struct vnode *vp;
3810 	int error;
3811 
3812 	/*
3813 	 * Must be super user
3814 	 */
3815 	error = suser(td);
3816 	if (error)
3817 		return (error);
3818 
3819 	error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
3820 	if (error)
3821 		return (error);
3822 
3823 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3824 		return (ESTALE);
3825 	if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)))
3826 		return (error);
3827 	error = vn_stat(vp, &sb, td->td_proc->p_ucred);
3828 	vput(vp);
3829 	if (error)
3830 		return (error);
3831 	error = copyout(&sb, uap->sb, sizeof(sb));
3832 	return (error);
3833 }
3834 
3835 /*
3836  * fhstatfs_args(struct fhandle *u_fhp, struct statfs *buf)
3837  */
3838 int
3839 sys_fhstatfs(struct fhstatfs_args *uap)
3840 {
3841 	struct thread *td = curthread;
3842 	struct proc *p = td->td_proc;
3843 	struct statfs *sp;
3844 	struct mount *mp;
3845 	struct vnode *vp;
3846 	struct statfs sb;
3847 	char *fullpath, *freepath;
3848 	fhandle_t fh;
3849 	int error;
3850 
3851 	/*
3852 	 * Must be super user
3853 	 */
3854 	if ((error = suser(td)))
3855 		return (error);
3856 
3857 	if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
3858 		return (error);
3859 
3860 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3861 		return (ESTALE);
3862 
3863 	if (p != NULL && !chroot_visible_mnt(mp, p))
3864 		return (ESTALE);
3865 
3866 	if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)))
3867 		return (error);
3868 	mp = vp->v_mount;
3869 	sp = &mp->mnt_stat;
3870 	vput(vp);
3871 	if ((error = VFS_STATFS(mp, sp, p->p_ucred)) != 0)
3872 		return (error);
3873 
3874 	error = mount_path(p, mp, &fullpath, &freepath);
3875 	if (error)
3876 		return(error);
3877 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3878 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
3879 	kfree(freepath, M_TEMP);
3880 
3881 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
3882 	if (suser(td)) {
3883 		bcopy(sp, &sb, sizeof(sb));
3884 		sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0;
3885 		sp = &sb;
3886 	}
3887 	return (copyout(sp, uap->buf, sizeof(*sp)));
3888 }
3889 
3890 /*
3891  * fhstatvfs_args(struct fhandle *u_fhp, struct statvfs *buf)
3892  */
3893 int
3894 sys_fhstatvfs(struct fhstatvfs_args *uap)
3895 {
3896 	struct thread *td = curthread;
3897 	struct proc *p = td->td_proc;
3898 	struct statvfs *sp;
3899 	struct mount *mp;
3900 	struct vnode *vp;
3901 	fhandle_t fh;
3902 	int error;
3903 
3904 	/*
3905 	 * Must be super user
3906 	 */
3907 	if ((error = suser(td)))
3908 		return (error);
3909 
3910 	if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
3911 		return (error);
3912 
3913 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3914 		return (ESTALE);
3915 
3916 	if (p != NULL && !chroot_visible_mnt(mp, p))
3917 		return (ESTALE);
3918 
3919 	if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)))
3920 		return (error);
3921 	mp = vp->v_mount;
3922 	sp = &mp->mnt_vstat;
3923 	vput(vp);
3924 	if ((error = VFS_STATVFS(mp, sp, p->p_ucred)) != 0)
3925 		return (error);
3926 
3927 	sp->f_flag = 0;
3928 	if (mp->mnt_flag & MNT_RDONLY)
3929 		sp->f_flag |= ST_RDONLY;
3930 	if (mp->mnt_flag & MNT_NOSUID)
3931 		sp->f_flag |= ST_NOSUID;
3932 
3933 	return (copyout(sp, uap->buf, sizeof(*sp)));
3934 }
3935 
3936 
3937 /*
3938  * Syscall to push extended attribute configuration information into the
3939  * VFS.  Accepts a path, which it converts to a mountpoint, as well as
3940  * a command (int cmd), and attribute name and misc data.  For now, the
3941  * attribute name is left in userspace for consumption by the VFS_op.
3942  * It will probably be changed to be copied into sysspace by the
3943  * syscall in the future, once issues with various consumers of the
3944  * attribute code have raised their hands.
3945  *
3946  * Currently this is used only by UFS Extended Attributes.
3947  */
3948 int
3949 sys_extattrctl(struct extattrctl_args *uap)
3950 {
3951 	struct nlookupdata nd;
3952 	struct mount *mp;
3953 	struct vnode *vp;
3954 	int error;
3955 
3956 	vp = NULL;
3957 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3958 	if (error == 0)
3959 		error = nlookup(&nd);
3960 	if (error == 0) {
3961 		mp = nd.nl_nch.mount;
3962 		error = VFS_EXTATTRCTL(mp, uap->cmd,
3963 				uap->attrname, uap->arg,
3964 				nd.nl_cred);
3965 	}
3966 	nlookup_done(&nd);
3967 	return (error);
3968 }
3969 
3970 /*
3971  * Syscall to set a named extended attribute on a file or directory.
3972  * Accepts attribute name, and a uio structure pointing to the data to set.
3973  * The uio is consumed in the style of writev().  The real work happens
3974  * in VOP_SETEXTATTR().
3975  */
3976 int
3977 sys_extattr_set_file(struct extattr_set_file_args *uap)
3978 {
3979 	char attrname[EXTATTR_MAXNAMELEN];
3980 	struct iovec aiov[UIO_SMALLIOV];
3981 	struct iovec *needfree;
3982 	struct nlookupdata nd;
3983 	struct iovec *iov;
3984 	struct vnode *vp;
3985 	struct uio auio;
3986 	u_int iovlen;
3987 	u_int cnt;
3988 	int error;
3989 	int i;
3990 
3991 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
3992 	if (error)
3993 		return (error);
3994 
3995 	vp = NULL;
3996 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3997 	if (error == 0)
3998 		error = nlookup(&nd);
3999 	if (error == 0)
4000 		error = ncp_writechk(&nd.nl_nch);
4001 	if (error == 0)
4002 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4003 	if (error) {
4004 		nlookup_done(&nd);
4005 		return (error);
4006 	}
4007 
4008 	needfree = NULL;
4009 	iovlen = uap->iovcnt * sizeof(struct iovec);
4010 	if (uap->iovcnt > UIO_SMALLIOV) {
4011 		if (uap->iovcnt > UIO_MAXIOV) {
4012 			error = EINVAL;
4013 			goto done;
4014 		}
4015 		MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
4016 		needfree = iov;
4017 	} else {
4018 		iov = aiov;
4019 	}
4020 	auio.uio_iov = iov;
4021 	auio.uio_iovcnt = uap->iovcnt;
4022 	auio.uio_rw = UIO_WRITE;
4023 	auio.uio_segflg = UIO_USERSPACE;
4024 	auio.uio_td = nd.nl_td;
4025 	auio.uio_offset = 0;
4026 	if ((error = copyin(uap->iovp, iov, iovlen)))
4027 		goto done;
4028 	auio.uio_resid = 0;
4029 	for (i = 0; i < uap->iovcnt; i++) {
4030 		if (iov->iov_len > INT_MAX - auio.uio_resid) {
4031 			error = EINVAL;
4032 			goto done;
4033 		}
4034 		auio.uio_resid += iov->iov_len;
4035 		iov++;
4036 	}
4037 	cnt = auio.uio_resid;
4038 	error = VOP_SETEXTATTR(vp, attrname, &auio, nd.nl_cred);
4039 	cnt -= auio.uio_resid;
4040 	uap->sysmsg_result = cnt;
4041 done:
4042 	vput(vp);
4043 	nlookup_done(&nd);
4044 	if (needfree)
4045 		FREE(needfree, M_IOV);
4046 	return (error);
4047 }
4048 
4049 /*
4050  * Syscall to get a named extended attribute on a file or directory.
4051  * Accepts attribute name, and a uio structure pointing to a buffer for the
4052  * data.  The uio is consumed in the style of readv().  The real work
4053  * happens in VOP_GETEXTATTR();
4054  */
4055 int
4056 sys_extattr_get_file(struct extattr_get_file_args *uap)
4057 {
4058 	char attrname[EXTATTR_MAXNAMELEN];
4059 	struct iovec aiov[UIO_SMALLIOV];
4060 	struct iovec *needfree;
4061 	struct nlookupdata nd;
4062 	struct iovec *iov;
4063 	struct vnode *vp;
4064 	struct uio auio;
4065 	u_int iovlen;
4066 	u_int cnt;
4067 	int error;
4068 	int i;
4069 
4070 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4071 	if (error)
4072 		return (error);
4073 
4074 	vp = NULL;
4075 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4076 	if (error == 0)
4077 		error = nlookup(&nd);
4078 	if (error == 0)
4079 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4080 	if (error) {
4081 		nlookup_done(&nd);
4082 		return (error);
4083 	}
4084 
4085 	iovlen = uap->iovcnt * sizeof (struct iovec);
4086 	needfree = NULL;
4087 	if (uap->iovcnt > UIO_SMALLIOV) {
4088 		if (uap->iovcnt > UIO_MAXIOV) {
4089 			error = EINVAL;
4090 			goto done;
4091 		}
4092 		MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
4093 		needfree = iov;
4094 	} else {
4095 		iov = aiov;
4096 	}
4097 	auio.uio_iov = iov;
4098 	auio.uio_iovcnt = uap->iovcnt;
4099 	auio.uio_rw = UIO_READ;
4100 	auio.uio_segflg = UIO_USERSPACE;
4101 	auio.uio_td = nd.nl_td;
4102 	auio.uio_offset = 0;
4103 	if ((error = copyin(uap->iovp, iov, iovlen)))
4104 		goto done;
4105 	auio.uio_resid = 0;
4106 	for (i = 0; i < uap->iovcnt; i++) {
4107 		if (iov->iov_len > INT_MAX - auio.uio_resid) {
4108 			error = EINVAL;
4109 			goto done;
4110 		}
4111 		auio.uio_resid += iov->iov_len;
4112 		iov++;
4113 	}
4114 	cnt = auio.uio_resid;
4115 	error = VOP_GETEXTATTR(vp, attrname, &auio, nd.nl_cred);
4116 	cnt -= auio.uio_resid;
4117 	uap->sysmsg_result = cnt;
4118 done:
4119 	vput(vp);
4120 	nlookup_done(&nd);
4121 	if (needfree)
4122 		FREE(needfree, M_IOV);
4123 	return(error);
4124 }
4125 
4126 /*
4127  * Syscall to delete a named extended attribute from a file or directory.
4128  * Accepts attribute name.  The real work happens in VOP_SETEXTATTR().
4129  */
4130 int
4131 sys_extattr_delete_file(struct extattr_delete_file_args *uap)
4132 {
4133 	char attrname[EXTATTR_MAXNAMELEN];
4134 	struct nlookupdata nd;
4135 	struct vnode *vp;
4136 	int error;
4137 
4138 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4139 	if (error)
4140 		return(error);
4141 
4142 	vp = NULL;
4143 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4144 	if (error == 0)
4145 		error = nlookup(&nd);
4146 	if (error == 0)
4147 		error = ncp_writechk(&nd.nl_nch);
4148 	if (error == 0)
4149 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4150 	if (error) {
4151 		nlookup_done(&nd);
4152 		return (error);
4153 	}
4154 
4155 	error = VOP_SETEXTATTR(vp, attrname, NULL, nd.nl_cred);
4156 	vput(vp);
4157 	nlookup_done(&nd);
4158 	return(error);
4159 }
4160 
4161 /*
4162  * Determine if the mount is visible to the process.
4163  */
4164 static int
4165 chroot_visible_mnt(struct mount *mp, struct proc *p)
4166 {
4167 	struct nchandle nch;
4168 
4169 	/*
4170 	 * Traverse from the mount point upwards.  If we hit the process
4171 	 * root then the mount point is visible to the process.
4172 	 */
4173 	nch = mp->mnt_ncmountpt;
4174 	while (nch.ncp) {
4175 		if (nch.mount == p->p_fd->fd_nrdir.mount &&
4176 		    nch.ncp == p->p_fd->fd_nrdir.ncp) {
4177 			return(1);
4178 		}
4179 		if (nch.ncp == nch.mount->mnt_ncmountpt.ncp) {
4180 			nch = nch.mount->mnt_ncmounton;
4181 		} else {
4182 			nch.ncp = nch.ncp->nc_parent;
4183 		}
4184 	}
4185 
4186 	/*
4187 	 * If the mount point is not visible to the process, but the
4188 	 * process root is in a subdirectory of the mount, return
4189 	 * TRUE anyway.
4190 	 */
4191 	if (p->p_fd->fd_nrdir.mount == mp)
4192 		return(1);
4193 
4194 	return(0);
4195 }
4196 
4197